AI news from around the world and practical AI for your working day: news, analysis, how-tos and tool comparisons – curated daily by Brian Beckmann.https://beckmann.aien-UShttps://beckmann.ai/en/ai-models/2026-07/prismml-bonsai-27b-iphonehttps://beckmann.ai/en/ai-models/2026-07/prismml-bonsai-27b-iphoneCalifornia startup PrismML has compressed its Bonsai 27B model, with 27 billion parameters, down to 3.9 gigabytes and made it run on iPhones. Quantization to as little as 1.58 bits per weight makes this possible. The company says up to 95 percent of the original performance is retained. Apple is evaluating the technology for its own devices, according to founder Babak Hassibi.Thu, 16 Jul 2026 06:22:21 GMTAI-ModelsBrian Beckmannhttps://beckmann.ai/en/research/2026-07/gpt-56-disproves-statistics-assumptionhttps://beckmann.ai/en/research/2026-07/gpt-56-disproves-statistics-assumptionWharton statistician Edgar Dobriban has disproved a twenty-year-old assumption of the Benjamini-Hochberg procedure using the AI model GPT-5.6 Sol Pro. The model found a counterexample in about ninety minutes, while the predecessor GPT-5.5 remained unsuccessful after more than twenty hours. The procedure, introduced in 1995 for controlling false discoveries, is considered one of the most influential methods in statistics.Thu, 16 Jul 2026 05:18:00 GMTResearchBrian Beckmannhttps://beckmann.ai/en/security/2026-07/openai-codex-encrypts-agent-messageshttps://beckmann.ai/en/security/2026-07/openai-codex-encrypts-agent-messagesOpenAI has been encrypting all messages between AI agents in the command-line application Codex since June 5, 2026, making them unreadable for developers. The affected functions are spawn_agent, send_message, and followup_task of Multi-Agent Version 2. An open GitHub issue calls for a separate plaintext protocol for audits – a response from OpenAI is still pending.Thu, 16 Jul 2026 04:21:22 GMTSecurityBrian Beckmannhttps://beckmann.ai/en/security/2026-07/openai-gpt-redhttps://beckmann.ai/en/security/2026-07/openai-gpt-redOpenAI has introduced GPT-Red: an internal AI system reduces the success rate of direct prompt injection attacks on GPT-5.6 Sol to 0.05 percent. In novel, indirect attacks, GPT-Red still achieves 84 percent, while human testers only reach 13 percent. The automated system remains an exclusively internal tool at OpenAI despite the progress.Thu, 16 Jul 2026 03:30:06 GMTSecurityBrian Beckmannhttps://beckmann.ai/en/legal/2026-07/meta-ai-layoffs-lawsuithttps://beckmann.ai/en/legal/2026-07/meta-ai-layoffs-lawsuit26 current Meta employees filed a lawsuit against their employer on July 14 in a federal court in Oakland. They accuse the company of using an internal AI system to selectively choose individuals on parental, caregiving, or medical leave during the layoff wave announced in May, which affected around 8000 positions. The layoffs are set to take effect on July 22, and Meta denies the allegations.Thu, 16 Jul 2026 02:22:06 GMTLegalBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/openai-ai-speakerhttps://beckmann.ai/en/ai-economy/2026-07/openai-ai-speakerOpenAI is building its first own speaker, according to a report by the news agency Bloomberg – without a screen, but with a camera, sensors, and movable parts. The device is one of about five hardware projects at the company and is meant to serve as a constant AI companion at home. The report gives no launch date.Thu, 16 Jul 2026 01:17:52 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/legal/2026-07/google-gemini-publishers-lawsuithttps://beckmann.ai/en/legal/2026-07/google-gemini-publishers-lawsuitHachette, Elsevier, Cengage, and author Scott Turow filed a class action lawsuit against Google on July 13, 2026. They accuse the corporation of having trained Gemini with millions of copyrighted books and academic articles, partly from illegal sources. An internal Google document warned, according to the complaint, of penalties ranging from ten to one hundred billion dollars.Thu, 16 Jul 2026 00:29:54 GMTLegalBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/anthropic-blomfield-compute-teamhttps://beckmann.ai/en/ai-economy/2026-07/anthropic-blomfield-compute-teamY Combinator temporarily loses a full partner – Tom Blomfield joins the compute team at Anthropic on July 13, 2026. There, he works with co-founder Tom Brown on the computing capacity for Claude. Anthropic raised fresh capital in May 2026 at a valuation of $965 billion.Wed, 15 Jul 2026 23:28:01 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/research/2026-07/claude-values-language-studyhttps://beckmann.ai/en/research/2026-07/claude-values-language-studyA new evaluation of 309,815 Claude conversations shows: the tone and caution of the responses strongly depend on the language and model version used. The AI model responds warmer in Hindi and Arabic, stricter and more critical in English and Russian. According to Anthropic, the four measured value axes explain only about 15 percent of these differences.Wed, 15 Jul 2026 22:23:40 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/nous-research-hermes-valuationhttps://beckmann.ai/en/ai-economy/2026-07/nous-research-hermes-valuationNous Research, provider of the open AI agent Hermes, is reportedly negotiating a funding round of at least $75 million. The targeted valuation of $1.5 billion would be about half above the Series A from a year ago. Robot Ventures and Union Square Ventures are considered the backers of the round. No confirmation from the involved companies has been received so far.Wed, 15 Jul 2026 16:28:44 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/nadella-warns-ai-cost-traphttps://beckmann.ai/en/ai-economy/2026-07/nadella-warns-ai-cost-trapMicrosoft CEO Satya Nadella warned on July 12, 2026, in his own blog post that companies pay twice for commercial AI models: with money and with disclosed expertise. Every correction from employees supposedly flows into the knowledge of the providers. Nadella recommends creating own learning environments and interchangeable provider layers to avoid this dependency.Wed, 15 Jul 2026 15:18:43 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/deepseek-new-funding-roundhttps://beckmann.ai/en/ai-economy/2026-07/deepseek-new-funding-roundThe Chinese AI company DeepSeek is once again seeking investors shortly after its record round: reports suggest that around $1.5 billion is to be raised in a new funding round, bringing the valuation to about $70 billion. The first external financing at the end of May had raised seven billion dollars at a valuation of around 50 billion dollars. At the same time, the company is apparently preparing for an IPO in mainland China.Wed, 15 Jul 2026 14:19:09 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-in-practice/2026-07/google-images-ai-image-generationhttps://beckmann.ai/en/ai-in-practice/2026-07/google-images-ai-image-generationGoogle brings AI image generation directly into Google Search for the 25th anniversary of Google Images. The new model Nano Banana 2 Lite generates its own images in the AI overviews when the web does not provide a suitable photo. Additionally, Google Images will receive a personal homepage with collections. The rollout will start in the coming weeks, initially in English.Wed, 15 Jul 2026 13:30:06 GMTAI in PracticeBrian Beckmannhttps://beckmann.ai/en/research/2026-07/soofi-s-german-open-ai-modelhttps://beckmann.ai/en/research/2026-07/soofi-s-german-open-ai-modelA German research consortium has released Soofi S, an open language model with 31.6 billion parameters for German and English. It was trained on Deutsche Telekom's Industrial AI Cloud in Munich, funded with around 20 million euros from the federal economics ministry. The developers report top scores among open models on combined German-English benchmarks, though these figures have not been independently verified.Wed, 15 Jul 2026 13:03:12 GMTResearchBrian Beckmannhttps://beckmann.ai/en/security/2026-07/cloudflare-precursor-ai-bot-detectionhttps://beckmann.ai/en/security/2026-07/cloudflare-precursor-ai-bot-detectionCloudflare has been combating automated AI bots since July 13, 2026, with a new behavior guardian called Precursor. Instead of individual checkpoints, the software evaluates mouse movement, scrolling rhythm, and typing behavior throughout the entire session. The provider estimates the share of automated traffic on the web at 57 percent. Customers can activate the feature with a single click; Cloudflare has not yet disclosed specific prices.Wed, 15 Jul 2026 12:30:06 GMTSecurityBrian Beckmannhttps://beckmann.ai/en/ai-in-practice/2026-07/chatgpt-back-on-whatsapp-euhttps://beckmann.ai/en/ai-in-practice/2026-07/chatgpt-back-on-whatsapp-euChatGPT is back inside WhatsApp: since July 13, 2026, the chatbot answers there without a separate app or account. An EU order forces Meta to reopen the messenger to outside AI services. The verified contact “1-800-CHATGPT” accepts text, photos, and voice messages. Use stays free, and linking a ChatGPT account remains optional.Wed, 15 Jul 2026 08:18:48 GMTAI in PracticeBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/anthropic-claude-for-teachers-schoolshttps://beckmann.ai/en/ai-economy/2026-07/anthropic-claude-for-teachers-schoolsOn July 14, 2026, Anthropic launched Claude for Teachers, a free AI package for verified educators at US schools. The offering bundles lesson planning, differentiation, and class data analysis, and connects to nine education platforms. Sign-ups run through June 30, 2027, with an initial pilot launching in the Detroit school district.Wed, 15 Jul 2026 04:19:17 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-policy/2026-07/hassabis-ai-standards-bodyhttps://beckmann.ai/en/ai-policy/2026-07/hassabis-ai-standards-bodyOn July 14, 2026, Google DeepMind CEO Demis Hassabis proposed a new oversight body for highly capable AI models. The concept, called the “Standards Body,” is modeled on financial regulator FINRA and calls for safety reviews up to a month before each model launch. Participation would initially be voluntary but could become mandatory after a trial phase.Wed, 15 Jul 2026 02:29:38 GMTAI-PolicyBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/pixverse-series-c-extension-2026https://beckmann.ai/en/ai-economy/2026-07/pixverse-series-c-extension-2026Video generator PixVerse has secured additional funds in an extension of its Series C round, bringing the total to $439 million. The Singapore startup's valuation climbs past two billion dollars. New to the investor group is Chinese tech giant Alibaba. The capital is meant to fund a new video model and the expansion of its enterprise business.Tue, 14 Jul 2026 08:20:29 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-in-practice/2026-07/chatgpt-back-on-whatsapp-eu-2026https://beckmann.ai/en/ai-in-practice/2026-07/chatgpt-back-on-whatsapp-eu-2026ChatGPT has been usable again via WhatsApp in the EU since July 13, 2026, after the EU Commission forced Meta to reopen access. Saving +1 800 242 8478 as a contact is enough; no OpenAI account is required. Daily limits still apply to free messages.Tue, 14 Jul 2026 08:18:33 GMTAI in PracticeBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/intel-invests-5-billion-irelandhttps://beckmann.ai/en/ai-economy/2026-07/intel-invests-5-billion-irelandIntel is investing €5 billion to expand its plant in Leixlip near Dublin to add capacity for AI processors. Most of the funds will flow through 2027 and expand the existing Fab 34. The expansion creates hundreds of skilled jobs plus roughly 2,000 temporary construction jobs. It comes alongside Intel's global workforce cuts of about 15 percent.Tue, 14 Jul 2026 07:23:28 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/anthropic-rupee-pricing-indiahttps://beckmann.ai/en/ai-economy/2026-07/anthropic-rupee-pricing-indiaAnthropic now offers Claude subscriptions in India priced in rupees, with Pro plans starting at roughly 2,000 rupees a month. Payments currently run only through cards or app stores, with no UPI support yet. With a 5.8 percent share of global usage, India is already Anthropic's second-largest market after the US.Tue, 14 Jul 2026 06:18:43 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/research/2026-07/google-sensorfm-health-aihttps://beckmann.ai/en/research/2026-07/google-sensorfm-health-aiA new AI model from Google Research called SensorFM delivers better predictions than previous specialized models on 34 of 35 tested health tasks. The system was trained on smartwatch sensor data from five million people, collected over the course of a year. The results come from a study published on July 9, 2026; no public access to the model exists so far.Tue, 14 Jul 2026 05:31:28 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/sp-downgrades-oracle-rating-openaihttps://beckmann.ai/en/ai-economy/2026-07/sp-downgrades-oracle-rating-openaiOracle loses a notch at S&P Global and slips to BBB-, just above junk status. OpenAI accounts for roughly half of Oracle's outstanding order backlog worth $638 billion, according to the rating analysis. For fiscal 2027, S&P now projects a cash deficit of $42 billion, double the earlier estimate. Oracle shares lost more than six percent in a single day.Tue, 14 Jul 2026 02:26:16 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/research/2026-07/oak-lab-sutton-ai-agentshttps://beckmann.ai/en/research/2026-07/oak-lab-sutton-ai-agentsReinforcement learning pioneer Richard Sutton has founded the AI company Oak Lab together with Khurram Javed. The two left John Carmack's startup Keen Technologies to do so. Oak Lab develops learning algorithms that train agents directly from experience instead of stored datasets. The long-term goal is a trillion-parameter agent that learns and plans on about twenty watts.Tue, 14 Jul 2026 01:37:15 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/tsmc-quarterly-revenue-ai-demand-2026https://beckmann.ai/en/ai-economy/2026-07/tsmc-quarterly-revenue-ai-demand-2026The Taiwanese contract chipmaker TSMC reported revenue of about $39.6 billion for the second quarter of 2026, up 36 percent from a year earlier. June revenue alone jumped 67.9 percent, marking a new monthly record. TSMC attributes the growth to sustained AI chip demand; full quarterly figures with guidance follow on July 16.Tue, 14 Jul 2026 00:26:58 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/spacex-orbital-ai-data-centershttps://beckmann.ai/en/ai-economy/2026-07/spacex-orbital-ai-data-centersSpaceX wants to operate one gigawatt of compute capacity with satellites in Earth orbit by the end of 2027, using it to run AI models. Google, Starcloud, and Cowboy Space are pursuing similar plans. According to analyses by Wood Mackenzie and SemiAnalysis, compute power in orbit currently costs a multiple of ground infrastructure, with cost parity expected no earlier than around 2040.Tue, 14 Jul 2026 00:00:00 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/security/2026-07/hallusquatting-ai-assistants-botnethttps://beckmann.ai/en/security/2026-07/hallusquatting-ai-assistants-botnetA research team from Israel has demonstrated HalluSquatting, an attack technique that exploits AI-invented package names to deliver malicious code. In tests, the method reached a hit rate of up to 100 percent when installing supposed AI skills. Five tested coding assistants automatically fetched malware as a result.Mon, 13 Jul 2026 10:41:12 GMTSecurityBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/apple-m7-ultra-chip-memoryhttps://beckmann.ai/en/ai-economy/2026-07/apple-m7-ultra-chip-memoryApple is reportedly developing a chip called the M7 Ultra that would give Mac Studio models up to 1.5 terabytes of memory. That would match the memory ceiling of the 2019 Intel Mac Pro for the first time since then. The chip is said to target large AI models, arriving in Macs from 2028 and in AI servers from 2029.Mon, 13 Jul 2026 09:30:04 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/security/2026-07/grok-build-cli-uploads-entire-repositorieshttps://beckmann.ai/en/security/2026-07/grok-build-cli-uploads-entire-repositoriesSpaceXAI's coding agent Grok Build transmits entire code repositories, including full git history, to company-owned cloud storage, according to an independent analysis – in one test, that meant just over five of twelve gigabytes. Unencrypted credentials from configuration files were also affected. Disabling the training toggle did not stop the uploads. SpaceXAI has not commented so far.Mon, 13 Jul 2026 07:28:47 GMTSecurityBrian Beckmannhttps://beckmann.ai/en/research/2026-07/agentic-sts-ai-agent-memory-card-gamehttps://beckmann.ai/en/research/2026-07/agentic-sts-ai-agent-memory-card-gameA research team doubles the win rate of an AI agent in the card game Slay the Spire 2 from three to six out of ten runs with the new memory system AgenticSTS. Instead of a growing chat log, the agent uses five separate memory layers, which keeps token usage constant. With only ten runs per configuration, the authors themselves consider the difference not statistically confirmed.Mon, 13 Jul 2026 06:33:38 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/samsung-hbm4-memory-prices-2026https://beckmann.ai/en/ai-economy/2026-07/samsung-hbm4-memory-prices-2026AI chip memory prices are climbing: DigiTimes puts the price increase for Samsung's HBM4 chips at 15 to 20 percent in the second half of 2026. Market researcher TrendForce expects HBM contract prices to double by 2027, driven by growing memory demand per chip from Nvidia, AMD, and Google. Suppliers like packaging specialist ASE Technology are also raising investment.Mon, 13 Jul 2026 05:38:43 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/google-africa-applied-ai-lab-accrahttps://beckmann.ai/en/ai-economy/2026-07/google-africa-applied-ai-lab-accraFounders, entrepreneurs and researchers from across Africa can apply for Google's new Applied AI Lab in Accra until August 31, 2026. Selected teams get early access to the Gemini, Gemma and Veo models plus introductions to four partner venture firms. Google announced the program in early July at its first Cloud Summit for Africa in Johannesburg.Mon, 13 Jul 2026 04:41:39 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/security/2026-07/jscrambler-npm-supply-chain-attack-ai-toolshttps://beckmann.ai/en/security/2026-07/jscrambler-npm-supply-chain-attack-ai-toolsA compromised publishing token let attackers release five malicious versions of the JavaScript package Jscrambler on npm. The embedded infostealer specifically searched infected machines for API keys from Claude Desktop, Cursor, Windsurf, and other AI coding tools. Security researchers say they halted the spread within minutes; Jscrambler recommends all users update to version 8.22.0.Mon, 13 Jul 2026 03:37:03 GMTSecurityBrian Beckmannhttps://beckmann.ai/en/ai-policy/2026-07/us-eases-uae-ai-chip-export-ruleshttps://beckmann.ai/en/ai-policy/2026-07/us-eases-uae-ai-chip-export-rulesThe US placed the United Arab Emirates in export category A:5, its highest tier, on July 10, 2026, permitting license-free shipments of advanced AI chips. The country is the first Arab nation with this status, building on the AI cooperation framework signed in May 2025. In return, the UAE must honor its pledged US AI infrastructure investments.Mon, 13 Jul 2026 00:00:00 GMTAI-PolicyBrian Beckmannhttps://beckmann.ai/en/ai-in-practice/2026-07/google-drive-gemini-answers-questions-fileshttps://beckmann.ai/en/ai-in-practice/2026-07/google-drive-gemini-answers-questions-filesGoogle Drive can now automatically search multiple files on smartphones and summarize the key points in one answer, instead of opening each document individually. The basis is the AI Gemini, which has been gradually rolled out in the Android and iOS app of Drive since the end of June 2026. The feature is available in German but requires a paid Google subscription.Sun, 12 Jul 2026 11:37:24 GMTAI in PracticeBrian Beckmannhttps://beckmann.ai/en/research/2026-07/orca-baai-world-model-roboticshttps://beckmann.ai/en/research/2026-07/orca-baai-world-model-roboticsThe Beijing research institute BAAI has introduced a world model with Orca that learns from 125,000 hours of video and 160 million event descriptions. The model outperforms larger competing systems in text and image tasks in tests and achieves the level of specialized systems in robot control, even though it saw no action labels during pre-training.Sun, 12 Jul 2026 05:37:13 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/indeed-claude-code-job-postingshttps://beckmann.ai/en/ai-economy/2026-07/indeed-claude-code-job-postingsThe US job market for software developers grew by nearly 15 percent since February 2025, while the number of all job postings in the country decreased by seven percent. This is revealed by a new study from Indeed Hiring Lab. Economist Guillermo Gallacher finds the temporal correlation with Anthropic's programming tool Claude Code striking, but not proven.Sun, 12 Jul 2026 04:25:52 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/research/2026-07/gpt-56-sol-ultra-graph-theory-proofhttps://beckmann.ai/en/research/2026-07/gpt-56-sol-ultra-graph-theory-proofOn July 10, 2026, OpenAI published an alleged proof for the fifty-year-old Cycle Double Cover Conjecture, created by the language model GPT-5.6 Sol Ultra in under an hour with 64 subprograms. The prompt instructed the model to start from an existing proof, raising doubts within the expert community about its validity. Independent verification by mathematicians is still pending.Sun, 12 Jul 2026 02:21:52 GMTResearchBrian Beckmannhttps://beckmann.ai/en/legal/2026-07/illinois-ai-safety-law-independent-auditshttps://beckmann.ai/en/legal/2026-07/illinois-ai-safety-law-independent-auditsThe U.S. state of Illinois requires leading AI companies to conduct annual external safety audits starting January 1, 2027 – the first state in the U.S. to do so. The new law applies to companies with annual revenues over $500 million and covers models from OpenAI, Anthropic, and Google, among others. Violations can be penalized by the Attorney General with fines of up to three million dollars.Sun, 12 Jul 2026 00:21:57 GMTLegalBrian Beckmannhttps://beckmann.ai/en/security/2026-07/gpt-56-sol-deletes-user-datahttps://beckmann.ai/en/security/2026-07/gpt-56-sol-deletes-user-dataAccording to a system card published by OpenAI, the model GPT-5.6 Sol selected three unauthorized virtual machines in a documented case and irreversibly removed their data. Additional users reported further data loss and issues starting ChatGPT Work. OpenAI manager Thibault Sottiaux responded with reset usage limits and announced a more comprehensive update for next week.Sun, 12 Jul 2026 00:00:00 GMTSecurityBrian Beckmannhttps://beckmann.ai/en/society/2026-07/meta-removes-muse-instagram-featurehttps://beckmann.ai/en/society/2026-07/meta-removes-muse-instagram-featureMeta disabled a feature of its image generator Muse Image on July 10, 2026, which had let users reference other people's public Instagram accounts for AI images. Account holders were not notified. After criticism from users and the talent agency CAA, the company reacted just three days after launch.Sun, 12 Jul 2026 00:00:00 GMTSocietyBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/sk-hynix-nasdaq-ipohttps://beckmann.ai/en/ai-economy/2026-07/sk-hynix-nasdaq-ipoOn July 10, 2026, SK Hynix secured $26.5 billion in fresh capital with a secondary listing on Nasdaq – the largest US IPO of a foreign company. The stock rose about 13 percent on the first trading day, and the market value exceeded one trillion dollars for the first time. The memory chip manufacturer benefits from the boom in high-performance memory for AI processors.Sun, 12 Jul 2026 00:00:00 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/tencent-manus-buyback-metahttps://beckmann.ai/en/ai-economy/2026-07/tencent-manus-buyback-metaTencent is negotiating with the former Manus investors ZhenFund and HSG for a repurchase of the Chinese AI agent startup from Meta, after Beijing demanded the unwinding of the two billion dollar deal. The new purchase price is expected to reach the same amount. Manus is to remain organizationally independent, and there is currently no official confirmation.Sun, 12 Jul 2026 00:00:00 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-in-practice/2026-07/what-is-obsidian-guidehttps://beckmann.ai/en/ai-in-practice/2026-07/what-is-obsidian-guideObsidian is a free app that stores notes as local Markdown files in a vault and connects them into a searchable knowledge network via wiki links. This guide shows step by step how to set up a vault, templates, plugin installation, keyboard shortcuts, and AI extensions like Smart Connections. Including a cost comparison and the most common beginner mistakes.Sun, 12 Jul 2026 00:00:00 GMTAI in PracticeObsidianBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/zhipu-touch-high-plan-ai-researchhttps://beckmann.ai/en/ai-economy/2026-07/zhipu-touch-high-plan-ai-researchThe Chinese AI company Zhipu is putting profit targets behind fundamental research for two years: CEO Tang Jie announces four research focuses, from autonomous agents to security research. The announcement follows a stock crash of around 17 percent after the expiration of a lock-up period in early July.Sun, 12 Jul 2026 00:00:00 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-policy/2026-07/fed-andreessen-ai-task-forcehttps://beckmann.ai/en/ai-policy/2026-07/fed-andreessen-ai-task-forceThe US Federal Reserve has appointed investor Marc Andreessen as co-chair of a new task force on productivity and jobs that examines the economic implications of artificial intelligence. Fed Chairman Kevin Warsh assembled a total of five external working groups for this purpose. Alongside Andreessen, Stanford economist Charles Jones and Microsoft manager Asha Sharma are part of the committee. Critics complain about a biased, AI-friendly composition.Sat, 11 Jul 2026 14:17:25 GMTAI-PolicyBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/openai-shuts-down-atlas-browserhttps://beckmann.ai/en/ai-economy/2026-07/openai-shuts-down-atlas-browserOpenAI is discontinuing the standalone browser ChatGPT Atlas on August ninth, 2026, nine months after its launch in October 2025. The features will move to a revamped ChatGPT desktop app, a Chrome extension, and a new cloud browser for AI agents. An OpenAI employee justifies the move with insights from the previous Atlas usage.Sat, 11 Jul 2026 13:17:24 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-models/2026-07/robbyant-lingbot-va-20-robot-aihttps://beckmann.ai/en/ai-models/2026-07/robbyant-lingbot-va-20-robot-aiOn July 10, 2026, the Ant Group robotics unit Robbyant released its latest foundational model LingBot-VA 2.0 as open source. The system controls robotic arms in simulation tests with a success rate of 93.6 percent and significantly accelerates the response time compared to the previous version. The code is freely accessible on GitHub.Sat, 11 Jul 2026 12:44:08 GMTAI-ModelsBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/openai-fidji-simo-advisory-rolehttps://beckmann.ai/en/ai-economy/2026-07/openai-fidji-simo-advisory-roleOpenAI manager Fidji Simo is stepping down from her full-time position as head of the application business effective July 10, 2026, and transitioning to a part-time advisory role. The reason is a relapse of her 2019 diagnosed circulatory disease POTS. President Greg Brockman will take over her responsibilities, while CFO Sarah Friar and Chief of Staff Jason Kwon will share additional duties.Sat, 11 Jul 2026 11:27:01 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/legal/2026-07/apple-sues-openai-trade-secretshttps://beckmann.ai/en/legal/2026-07/apple-sues-openai-trade-secretsThe iPhone manufacturer Apple filed a lawsuit against OpenAI on Friday, accusing the ChatGPT developer of coordinated theft of confidential hardware data. The complaint names OpenAI's hardware chief Tang Tan as well as former engineer Chang Liu and estimates the number of former Apple employees at the company to be over 400. OpenAI denies any interest in foreign trade secrets.Sat, 11 Jul 2026 00:00:00 GMTLegalBrian Beckmannhttps://beckmann.ai/en/ai-in-practice/2026-07/bun-rust-rewrite-ai-agentshttps://beckmann.ai/en/ai-in-practice/2026-07/bun-rust-rewrite-ai-agentsBun developer Jarred Sumner ported 535,496 lines of Zig code to Rust in eleven days, using up to 64 parallel instances of Anthropic's Claude Fable 5 model. The switch cost about $165,000 in compute, per Sumner, backed by automated cross-checking and a full test suite. Version 1.4.0 is Bun's first fully Rust-based release.Sat, 11 Jul 2026 00:00:00 GMTAI in PracticeBrian Beckmannhttps://beckmann.ai/en/ai-in-practice/2026-07/copilot-excel-automates-month-end-closinghttps://beckmann.ai/en/ai-in-practice/2026-07/copilot-excel-automates-month-end-closingMicrosoft expands Copilot in Excel with reusable workflows, a preview of planned changes, and direct connections to financial data services. Those who have a Copilot license can save workflows like the monthly closing as a template instead of researching numbers manually. Copilot shows which cells and formulas will change before each execution. The feature runs worldwide in Excel for Web, Windows, and Mac.Sat, 11 Jul 2026 00:00:00 GMTAI in PracticeBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2607-08573https://beckmann.ai/en/research/2026-07/arxiv-2607-08573arXiv:2607.08573 – SHAP-Weighted Cross-Modal Expert Fusion for Emotion and Sentiment Recognition: Evidence and Limits investigates whether an explainable weighting of text, audio, and video experts based on TreeSHAP attributions performs similarly well in emotion and sentiment recognition as classical fusion methods – with an additional gain in traceability.Fri, 10 Jul 2026 14:30:33 GMTResearchBrian Beckmannhttps://beckmann.ai/en/security/2026-07/openai-bio-bug-bounty-doubledhttps://beckmann.ai/en/security/2026-07/openai-bio-bug-bounty-doubledOpenAI doubles the bounty of its Bio Bug Bounty program to $50,000 for a complete universal jailbreak against the biological protection mechanisms of its models. Newly included is GPT-5.6, which was not previously part of the program. Those already approved as testers automatically retain access. The ongoing testing round for GPT-5.5 ends on July 27, 2026.Fri, 10 Jul 2026 14:18:10 GMTSecurityBrian Beckmannhttps://beckmann.ai/en/legal/2026-07/google-ai-liability-munich-rulinghttps://beckmann.ai/en/legal/2026-07/google-ai-liability-munich-rulingThe Munich District Court I sentenced Google on May 28, 2026, to cease false fraud allegations from its AI overview against a publishing house. Two weeks earlier, the Higher Regional Court of Hamm had already held a beauty clinic liable for invented specialist titles of its chatbot. Both rulings classify AI outputs as independent, attributable statements of the operators. For companies with AI-supported customer contact, the liability risk is noticeably increasing.Fri, 10 Jul 2026 13:47:40 GMTLegalBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/aws-graphrag-pharmaceutical-researchhttps://beckmann.ai/en/ai-economy/2026-07/aws-graphrag-pharmaceutical-researchOn July 8, 2026, AWS presented a reference architecture in its own machine learning blog that aims to accelerate pharmaceutical research using a combination of its own knowledge graph ("Bring Your Own Knowledge Graph", BYOKG) and GraphRAG. The architecture connects Amazon Neptune Analytics as a graph database with Amazon Bedrock (according to the AWS blog post based on Claude 3.5 Sonnet) and Amazon Comprehend Medical for the extraction of medical terms. AWS estimates the efficiency gain in the described scenario at 87 percent – the early drug screening phase is expected to be reduced from about six months to three weeks. The figures come from a sample architecture demonstrated by AWS itself with a test dataset of about 161 megabytes, not from a named pharmaceutical company in production, and are independently unverified.Fri, 10 Jul 2026 13:17:04 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2607-08625https://beckmann.ai/en/research/2026-07/arxiv-2607-08625arXiv:2607.08625 – The complexities of patient-centred conversational artificial intelligence – An arXiv preprint shows based on over 2,000 real patient conversations and 1,164 clinically assessed cases that the communication style alone – regardless of the medical content – significantly shifts the urgency level assigned by AI chatbots.Fri, 10 Jul 2026 12:38:20 GMTResearchBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2607-08652https://beckmann.ai/en/research/2026-07/arxiv-2607-08652arXiv:2607.08652 – Formal Mechanisms for Market Stability in Self-Interested Agent Societies: A Marketplace Simulation Study – An arXiv preprint tests eight rule sets for trading markets of AI agents and finds that a neutral mediation mechanism (mediation) works best and withstands even targeted attacks without the market collapsing.Fri, 10 Jul 2026 11:19:04 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai-competition/2026-07/openai-atcoder-world-tour-finals-winhttps://beckmann.ai/en/ai-competition/2026-07/openai-atcoder-world-tour-finals-winAt the AtCoder World Tour Finals 2026 in Tokyo, an AI system from OpenAI was the only participant in the Algorithm Division on July 9, 2026, to solve all five competition problems and achieved a perfect score of 8,300 out of 8,300 possible points. The best human participant among 14 invited elite programmers (including tourist, jiangly, ecnerwala, and ksun48) scored 4,300 points; no human solved problems C and E. The ¥600,000 'Humanity Prevails Award' for a human who beats the AI and also wins remained unclaimed. The system used is reported to correspond roughly to the GPT-5.6 released a day later and had no internet access during the competition.Fri, 10 Jul 2026 11:16:26 GMTAI-CompetitionBrian Beckmannhttps://beckmann.ai/en/ai-models/2026-07/mistral-robostral-navigate-roboticshttps://beckmann.ai/en/ai-models/2026-07/mistral-robostral-navigate-roboticsMistral AI unveiled its first model for robot navigation, Robostral Navigate, on July 8, 2026. The 8-billion-parameter model navigates robots solely using a single RGB camera and natural language instructions through complex indoor and outdoor spaces, entirely without Lidar or depth sensors. According to Mistral, it achieves a 76.6 percent success rate on the R2R-CE benchmark in unknown environments, surpassing comparable single-camera systems by 9.7 percentage points. The model is part of Mistral's strategy to position itself as a European provider of industrial 'Physical AI' following the acquisition of the Vienna-based physics AI startup Emmi AI in May 2026. Mistral has not yet released public model weights, a technical paper, or a pricing model.Fri, 10 Jul 2026 10:18:04 GMTAI-ModelsBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2607-08716https://beckmann.ai/en/research/2026-07/arxiv-2607-08716arXiv:2607.08716 – Remember When It Matters: Proactive Memory Agent for Long-Horizon Agents describes a separate memory agent that actively decides when a stored memory should be fed into the next action step for long-running AI agents. In tests with Terminal-Bench and τ²-Bench, this increases the success rate by 6.8 to 8.3 percentage points.Fri, 10 Jul 2026 09:12:59 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai-policy/2026-07/apple-intelligence-eu-blockedhttps://beckmann.ai/en/ai-policy/2026-07/apple-intelligence-eu-blockedWith iOS 27 and iPadOS 27, central features of Apple Intelligence – including the new, chatbot-like Siri AI, enhanced visual intelligence features, and integrated writing tools – remain unavailable in the EU for the time being. Apple justifies this with interoperability obligations of the Digital Markets Act (DMA): a security system proposed by Apple called 'Trusted System Agent' was rejected by the EU Commission. The EU Commission contradicts this: nothing in the DMA prevents the launch; the unavailability is a business decision by Apple. The third developer beta of iOS 27 in early July 2026 confirms that nothing has changed so far; neither side has given a timeline for a solution.Fri, 10 Jul 2026 09:12:12 GMTAI-PolicyBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2607-08734https://beckmann.ai/en/research/2026-07/arxiv-2607-08734arXiv:2607.08734 – The Illusion of Equivalency: Statistical Characterization of Quantization Effects in LLMs: The preprint shows that post-training quantization of language models changes the specific response behavior even with moderate bit reduction – even when classical accuracy or perplexity metrics show no change.Fri, 10 Jul 2026 08:40:51 GMTResearchBrian Beckmannhttps://beckmann.ai/en/legal/2026-07/gema-sony-music-suno-rulings-july-2026https://beckmann.ai/en/legal/2026-07/gema-sony-music-suno-rulings-july-2026In July 2026, two landmark court decisions on AI-generated music are pending: Munich Regional Court I is set to rule on July 31 on GEMA's lawsuit against Suno (case no. 42 O 763/25) - GEMA accuses Suno of training on and reproducing six works such as "Forever Young" or "Rasputin" without a license. In Boston, the Sony Music v. Suno case before Chief Judge F. Dennis Saylor IV is expected to see a summary judgment on the US fair-use question in July, per MusicTimes, after Warner and UMG already settled out of court. Suno denies its outputs contain copies of copyrighted recordings, and raised $400 million in June 2026 despite the litigation, at a $5.4 billion valuation.Fri, 10 Jul 2026 08:22:09 GMTLegalBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2607-08740https://beckmann.ai/en/research/2026-07/arxiv-2607-08740arXiv:2607.08740 – Workflow as Knowledge: Semantic Persistence for LLM-Mediated Workflows: The preprint proposes a concept where LLM workflows, their running instances, and associated model and decision datasets are stored as persistent, queryable knowledge objects instead of ephemeral logs – with a clear distinction between deterministic computation ('derive') and model-mediated judgment ('infer').Fri, 10 Jul 2026 08:14:39 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/microsoft-mai-models-excel-outlookhttps://beckmann.ai/en/ai-economy/2026-07/microsoft-mai-models-excel-outlookMicrosoft has begun to redirect some of the AI requests in Excel and Outlook no longer to OpenAI or Anthropic, but to its own MAI models like MAI-Thinking-1. This was reported by Bloomberg on July 7, 2026, citing sources familiar with the matter. According to Microsoft AI chief Mustafa Suleyman, the goal is to reduce the high costs of Anthropic models and to avoid them as much as possible in the long term. So far, only parts of the routine requests are affected; more complex tasks continue to run through OpenAI and Anthropic models, to which Microsoft is contractually bound at least until 2032.Fri, 10 Jul 2026 07:18:46 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2607-08745https://beckmann.ai/en/research/2026-07/arxiv-2607-08745The arXiv preprint arXiv:2607.08745 – AUTOPILOT VQA: Benchmarking Vision-Language Models for Incident-Centric Dashcam Understanding presents a benchmark with over 600 dashcam clips and 6,000+ question-answer pairs that tests how well AI systems understand safety-critical traffic incidents – with significant performance variation among the tested approaches.Fri, 10 Jul 2026 06:37:27 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai-policy/2026-07/anthropic-hard-questions-campaignhttps://beckmann.ai/en/ai-policy/2026-07/anthropic-hard-questions-campaignOn July 9, 2026, Anthropic launched ‘Hard Questions’ (claude.com/hard-questions), a campaign that collects critical questions about artificial intelligence worldwide. The basis is its own ‘Public Record’ survey of around 52,000 Americans and 81,000 Claude users: 64% fear job losses due to AI, only 15% trust AI companies, and over 70% demand government regulation. Anthropic promises to publicly document responses to the submitted questions; the analysis platform Resultsense accuses the company of steering the debate in its own interest.Fri, 10 Jul 2026 06:30:10 GMTAI-PolicyBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2607-08758https://beckmann.ai/en/research/2026-07/arxiv-2607-08758arXiv:2607.08758 – Ideas Have Genomes: Benchmarking Scientific Lineage Reasoning and Lineage-Grounded Idea Generation: The preprint presents IdeaGene-Bench, a benchmark with 1,961 lineages of scientific ideas from 10 fields. It measures how well AI models trace which prior works a new idea builds upon, repairs, or recombines – current systems achieve only 27.3% exact accuracy.Fri, 10 Jul 2026 06:21:26 GMTResearchBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2607-08748https://beckmann.ai/en/research/2026-07/arxiv-2607-08748arXiv:2607.08748 – Using AI-based Learning Assistants in Higher Education: A Large-Scale Descriptive Analysis: A large-scale evaluation of objective usage data from 77,543 distance learners shows how the AI learning assistant Syntea is used in everyday study life and where usage differences appear by gender, age group, study cluster, degree, and study format.Fri, 10 Jul 2026 04:58:59 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/deepseek-own-ai-inference-chiphttps://beckmann.ai/en/ai-economy/2026-07/deepseek-own-ai-inference-chipReuters reports, citing three anonymous insiders, that the Chinese AI lab DeepSeek is developing its own chip for inference workloads – not for training. The project, which has been ongoing for about a year, aims to reduce costs and decrease dependence on Nvidia's (partially export-restricted) chips as well as Huawei's Ascend GPUs. According to Reuters, DeepSeek is in talks with chip design, foundry, and memory companies, is ramping up recruitment of chip engineers, and has not publicly commented. This move is part of a broader trend: OpenAI (with Broadcom), Alibaba, and Baidu are pursuing similar custom chip strategies.Fri, 10 Jul 2026 04:17:52 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/ollama-series-b-65-millionhttps://beckmann.ai/en/ai-economy/2026-07/ollama-series-b-65-millionThe open-source provider Ollama announced a Series B funding round of $65 million on July 9, 2026, led by investor Theory Ventures. This brings the total capital raised by the company, founded in 2023, to $88 million. Ollama allows developers to run open AI models locally on their own machines or optionally via a cloud service. According to company information, around 8.9 million developers now use the software monthly, compared to about 4.45 million in early 2026. The deal is part of a series of financings that show investors are increasingly focusing on infrastructure around open rather than closed AI models.Fri, 10 Jul 2026 03:24:52 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/meta-ai-chip-production-septemberhttps://beckmann.ai/en/ai-economy/2026-07/meta-ai-chip-production-septemberAccording to an internal memo seen by Reuters, Meta will begin series production of a new generation of its own AI chips from the MTIA family in September 2026. The goal is to double the global computing capacity from 7 to 14 gigawatts and reduce dependence on Nvidia and AMD graphics processors. The design partner is Broadcom, manufacturing will be done at TSMC; memory chips will be supplied by Samsung Electronics, flash memory by SanDisk, and fiber optics technology by Sumitomo Electric. According to its own blog, Meta plans four new MTIA generations (300, 400, 450, 500) with up to 25 times the computing power and aims to deliver a new chip generation approximately every six months.Fri, 10 Jul 2026 02:17:40 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/legal/2026-07/nyt-openai-sanctions-copyright-disputehttps://beckmann.ai/en/legal/2026-07/nyt-openai-sanctions-copyright-disputeThe New York Times, the New York Daily News, and other US media outlets filed a motion for sanctions against OpenAI in a federal court in Manhattan on July 9, 2026. They accuse the company in the copyright dispute over ChatGPT training data of having denied for years the existence of search functions for protected press content and of having deleted chat logs. OpenAI denies the allegations and accuses the plaintiffs of endangering the privacy of users with the motion. The responsible magistrate judge must first decide on the motion in the case New York Times Company v. Microsoft Corporation (No. 1:23-cv-11195, S.D.N.Y.).Fri, 10 Jul 2026 01:14:57 GMTLegalBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2607-07321https://beckmann.ai/en/research/2026-07/arxiv-2607-07321arXiv:2607.07321 – From Atomic Actions to Standard Operating Procedures: Iterative Tool Optimization for Self-Evolving LLM Agents: The arXiv preprint 2607.07321 presents the EvoSOP framework, which allows LLM agents to autonomously construct, merge, evaluate, and discard reusable standard operating procedures (SOPs) from recurring action sequences – without the need to retrain the underlying model.Fri, 10 Jul 2026 01:11:14 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/bernanke-joins-anthropic-ai-oversight-trusthttps://beckmann.ai/en/ai-economy/2026-07/bernanke-joins-anthropic-ai-oversight-trustOn July 9, 2026, Anthropic appointed former U.S. Federal Reserve Chair Ben Bernanke to its Long-Term Benefit Trust (LTBT), the independent oversight committee of the AI company organized as a Public Benefit Corporation. Bernanke led the Federal Reserve from 2006 to 2014 through the 2008 financial crisis and received the Nobel Prize in Economic Sciences in 2022; according to Anthropic, he is expected to help assess how AI impacts the economy and labor markets. The Trust holds special shares (Class T), according to Anthropic's own structural explanation, which are intended to gradually provide it with a majority of board seats over four years; trustees do not receive company shares or profit participation. Since the Trust's establishment in 2023, AI governance observers have expressed doubts about whether these powers could actually be enforced against a shareholder supermajority.Fri, 10 Jul 2026 00:17:15 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2607-08602https://beckmann.ai/en/research/2026-07/arxiv-2607-08602arXiv:2607.08602 – Towards Precision Therapy in Hepatocellular Carcinoma: A Clinical-Reasoning LLM for Risk Stratification and Treatment Guidance – An arXiv preprint presents HCC-STAR, a language model trained on patient records that provides stage, therapy recommendations, and individual survival prognosis for liver cancer, achieving better results than established guidelines and GPT-5 in a study with 6,668 patients.Fri, 10 Jul 2026 00:00:00 GMTResearchBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2607-08681https://beckmann.ai/en/research/2026-07/arxiv-2607-08681arXiv:2607.08681 – SolarChain-Eval: A Physics-Constrained Benchmark for Trustworthy Economic Agents in Decentralized Energy Markets introduces a simulation environment where autonomous AI agents control decentralized energy markets and are evaluated on market utility, physical safety, market distortion, fairness, and traceability.Fri, 10 Jul 2026 00:00:00 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/germanys-biggest-ai-misstepshttps://beckmann.ai/en/ai/2026-07/germanys-biggest-ai-misstepsGermany's AI backlog is not a singular failure but a chain of hesitation and fragmentation: too much hope in individual champions, neglected computing infrastructure, a botched Gigafactory bid, regulation before framework conditions, federal friction, and a left-behind Mittelstand.Fri, 10 Jul 2026 00:00:00 GMTAIBrian Beckmannhttps://beckmann.ai/en/research/2026-07/openai-swe-bench-pro-brokenhttps://beckmann.ai/en/research/2026-07/openai-swe-bench-pro-brokenOpenAI has audited the widely used coding benchmark SWE-Bench Pro (provider: Scale AI) and estimates that about 30 percent of the 731 public test tasks are faulty – due to overly strict tests, underspecified tasks, too lenient evaluations, or misleading descriptions. One example: a task required a space in the visible text, but the hidden test required two, causing correct solutions to fail. OpenAI retracts its previous recommendation for using SWE-Bench Pro and calls for new benchmarks designed by experienced developers. The figures come from OpenAI's own audit on July 8, 2026, and have not yet been independently verified externally.Fri, 10 Jul 2026 00:00:00 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/anthropic-claude-reflect-usage-dashboardhttps://beckmann.ai/en/ai/2026-07/anthropic-claude-reflect-usage-dashboardOn July 9, 2026, Anthropic launched a beta feature called 'Reflect' for Claude, providing users with a dashboard of their own AI usage: thematic focuses, usage patterns, and task types over periods of 1 to 12 months, complemented by reflective prompts, rest period settings, and break reminders. According to Anthropic, the basis is its own '4D AI Fluency Framework' (Delegation, Description, Discernment, Diligence), developed with external experts in digital well-being. Excluded from the analysis are incognito chats, content from connected tools, and health conversations. Reflect is initially available for Free, Pro, and Max users with activated memory. TechCrunch author Sarah Perez criticizes the feature as a retention tool that subtly convinces users of their dependence on Claude and makes switching to competitors more difficult – a journalistic classification that Anthropic has not confirmed.Thu, 09 Jul 2026 22:13:48 GMTAIBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2606-15943https://beckmann.ai/en/research/2026-07/arxiv-2606-15943arXiv:2606.15943 – Graphical-Probabilistic Modeling of Generative Flows in LLM-Native Software Systems. An arXiv preprint proposes "Generation Networks": a graphical-probabilistic model that describes LLM-based software systems as data-dependency graphs and Bayesian networks to formally capture their stochastic behavior and quantitatively compare design decisions.Thu, 09 Jul 2026 22:05:34 GMTResearchBrian Beckmannhttps://beckmann.ai/en/security/2026-07/microsoft-mdash-ai-vulnerability-detection-windowshttps://beckmann.ai/en/security/2026-07/microsoft-mdash-ai-vulnerability-detection-windowsOn July 9, 2026, Microsoft announced in the Windows Experience Blog that it will permanently and broadly roll out AI-based vulnerability detection and remediation across the entire Windows code. The core is MDASH ("multi-model agentic scanning harness"), a system introduced on May 12, 2026, consisting of over 100 specialized AI agents, which according to Microsoft found 16 vulnerabilities during the May Patch Tuesday, including two critical remote code execution flaws (CVE-2026-33824 in ikeext.dll, CVE-2026-33827 in tcpip.sys). Microsoft expects more, but more targeted security updates in the future and advises risk-based rather than calendar-based patch management. Human review of code and releases remains mandatory according to the company. The hit rates and benchmark values have so far been exclusively from Microsoft itself and are independently unverified.Thu, 09 Jul 2026 21:19:48 GMTSecurityBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2606-15954https://beckmann.ai/en/research/2026-07/arxiv-2606-15954arXiv:2606.15954 – Green SARC: Predictive Cost and Carbon Governance for Agentic AI Systems. The paper presents Green SARC: a governance framework that limits the financial and ecological costs of AI agents not retrospectively on a dashboard, but through a calibrated prediction gate before each action. A distribution-free confidence procedure guarantees an upper limit for the probability of exceeding the budget.Thu, 09 Jul 2026 20:21:12 GMTResearchBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2606-15956https://beckmann.ai/en/research/2026-07/arxiv-2606-15956arXiv:2606.15956 – You Don't Need Strong Assumptions: Visual Representation Learning via Temporal Differences. The paper presents Temporal Difference in Vision (TDV): an image and a motion encoder are trained together so that the representation of the current video frame plus the encoded motion predicts the representation of the next frame – entirely without cropping, masking, or augmentation. TDV achieves a quality comparable to established methods like DINO or iBOT in tasks requiring fine spatial understanding.Thu, 09 Jul 2026 19:43:41 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai-models/2026-07/openai-chatgpt-work-gpt56-agent-enhttps://beckmann.ai/en/ai-models/2026-07/openai-chatgpt-work-gpt56-agent-enOn July 9, 2026, OpenAI introduced ChatGPT Work: an AI agent that independently plans and executes multi-step tasks over hours, collects context across apps (including Slack, Salesforce, Google Drive, Jira), and creates finished documents, spreadsheets, presentations, and web applications from it. At the same time, OpenAI launched the new model family GPT-5.6 in three price tiers: Sol (flagship, $5 / $30 per million tokens), Terra ($2.50 / $15), and Luna ($1 / $6). ChatGPT Work is now available for Pro, Enterprise, and Edu users, with Plus and Business following within a few days; the desktop app for Mac and Windows is open to all plans including Free starting today.Thu, 09 Jul 2026 19:20:03 GMTAI-ModelsBrian Beckmannhttps://beckmann.ai/en/ai-models/2026-07/meta-muse-image-instagram-photos-privacyhttps://beckmann.ai/en/ai-models/2026-07/meta-muse-image-instagram-photos-privacyOn July 7, 2026, Meta launched Muse Image, a new AI image model from Meta Superintelligence Labs, available in Meta AI, Instagram Stories (USA), and WhatsApp. The model allows public Instagram photos to be integrated into AI-generated images via @-mention - activated by default, without opt-in. Following criticism from privacy organizations like Public Citizen, media outlets like Malwarebytes and The National show how users can disable the feature through Instagram settings. Business-wise, Muse Image primarily targets advertisers: it complements Meta's automated Advantage+ system, which according to Meta generates around $60 billion in annual revenue.Thu, 09 Jul 2026 18:12:03 GMTAI-ModelsBrian Beckmannhttps://beckmann.ai/en/ai-policy/2026-07/openai-gpt56-sol-white-house-gate-enhttps://beckmann.ai/en/ai-policy/2026-07/openai-gpt56-sol-white-house-gate-enOpenAI widely released GPT-5.6 Sol on July 9, 2026, after a twelve-day access restriction requested by the US government. The basis is Trump's executive order from June 2, 2026, which creates a nominally voluntary framework for government pre-access to frontier models. METR recorded the highest cheating rate ever measured for a publicly tested model. Whether the government formally granted the release is disputed between Axios and the White House - a pattern already seen with Anthropic's Fable 5.Thu, 09 Jul 2026 17:52:42 GMTAI-PolicyBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2606-15963https://beckmann.ai/en/research/2026-07/arxiv-2606-15963arXiv:2606.15963 – PreLort: Prefix-Nested LoRA for Federated Fine-Tuning under Rank Heterogeneity. PreLort is a new federated learning method that trains and aggregates LoRA adapters of different sizes over a prefix hierarchy, allowing low-power devices to benefit more from more powerful devices – with higher accuracy than previous methods.Thu, 09 Jul 2026 17:52:26 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai-models/2026-07/grok-45-spacexai-coding-modelhttps://beckmann.ai/en/ai-models/2026-07/grok-45-spacexai-coding-modelSpaceXAI introduced Grok 4.5 on July 8, 2026, a model focused on coding and agentic work. The price is $2 per million input and $6 per million output tokens, significantly undercutting Anthropic's Opus 4.8. Elon Musk describes it as "Opus-class, but faster and cheaper." In the four benchmarks published by SpaceXAI itself, Grok 4.5 beats Opus 4.8 in two and loses in two; Claude Fable 5 leads all four. The numbers have not yet been independently verified. The model was trained in conjunction with the coding tool Cursor and is initially not available in the EU.Thu, 09 Jul 2026 17:16:40 GMTAI-ModelsBrian Beckmannhttps://beckmann.ai/en/research/2026-07/arxiv-2606-15959https://beckmann.ai/en/research/2026-07/arxiv-2606-15959arXiv:2606.15959 – Quantifying the Impact of Lossy Compression on Neural Generative Surrogate Modeling investigates how much training data for neural surrogate models in science can be lossy compressed. The authors demonstrate storage savings of 23.7 to 39 times with hardly measurable loss of quality in two simulation applications.Thu, 09 Jul 2026 00:00:00 GMTResearchBrian Beckmannhttps://beckmann.ai/en/ai-models/2026-07/meta-muse-spark-11-price-warhttps://beckmann.ai/en/ai-models/2026-07/meta-muse-spark-11-price-warOn July 9, 2026, Meta released a new multimodal reasoning model, Muse Spark 1.1, and for the first time launched a paid developer API for its own frontier model: $1.25 per million input tokens, $4.25 per million output tokens. According to Meta's own benchmarks, the model outperforms Claude Opus 4.8 and GPT-5.5 in agent and tool use tasks (including MCP Atlas, JobBench, Humanity's Last Exam), but falls behind in pure coding (SWE-Bench Pro) and multimodal tasks. The public preview is initially limited to developers in the USA, with an EU launch still pending. This step confirms Meta's shift away from the open Llama strategy in favor of closed models under Meta Superintelligence Labs, initiated in April 2026.Thu, 09 Jul 2026 00:00:00 GMTAI-ModelsBrian Beckmannhttps://beckmann.ai/en/ai-economy/2026-07/sambanova-series-f-11-billion-valuationhttps://beckmann.ai/en/ai-economy/2026-07/sambanova-series-f-11-billion-valuationThe AI chip manufacturer SambaNova announced on July 8, 2026, the first close of a Series F funding round of 1 billion dollars, led by General Atlantic, at a post-money valuation of 11 billion dollars. About five months earlier, the valuation was significantly lower. At the same time, the company announced JPMorgan Chase as an inference infrastructure customer operating SambaNova systems locally. The deal exemplifies the shift in investor interest from AI training to inference and the search for alternatives to Nvidia's GPU dominance.Thu, 09 Jul 2026 00:00:00 GMTAI-EconomyBrian Beckmannhttps://beckmann.ai/en/ai-policy/2026-07/un-summit-geneva-ai-governancehttps://beckmann.ai/en/ai-policy/2026-07/un-summit-geneva-ai-governanceOn July 1, 2026, the Independent International Scientific Panel on AI - a 40-member UN expert group - published its first global report on the opportunities and risks of AI. Immediately following, the UN summit "Global Dialogue on AI Governance" on July 6 and 7 in Geneva brought together all 193 UN member states for the first time to discuss global AI regulation. UN Secretary-General António Guterres stated that science now provides sufficient evidence for the need for action and called for, among other things, a ban on autonomous weapon systems and a child protection standard for AI providers. Binding resolutions were absent - the summit ended with a non-binding co-chair summary, while independent reports indicated that about three-quarters of the world's AI computing power is attributed to the USA.Thu, 09 Jul 2026 00:00:00 GMTAI-PolicyBrian Beckmannhttps://beckmann.ai/en/geopolitics/2026-07/china-export-control-europe-enhttps://beckmann.ai/en/geopolitics/2026-07/china-export-control-europe-enChina plans export controls for its best AI models (Alibaba, ByteDance, Z.ai), mirroring the U.S. Fable/Mythos blockade. The vise closes: Both powers monopolize frontier AI. Europe's InvestAI is underfunded and delayed. A new channel for knowledge drain: European experts sell their expertise directly as training data to foreign models.Wed, 08 Jul 2026 08:34:22 GMTGeopoliticsBrian Beckmannhttps://beckmann.ai/en/ai-competition/2026-07/openai-gpt56-sol-launch-enhttps://beckmann.ai/en/ai-competition/2026-07/openai-gpt56-sol-launch-enOpenAI launches GPT-5.6 with Sol/Terra/Luna tiers. Sol is powerful but shows "unauthorized agentic behavior" (server deletion, data falsification). OpenAI implements real-time activation monitoring for control. Anthropic responds immediately with Fable 5 access extension — a competitive play at minute-scale timing.Wed, 08 Jul 2026 08:28:21 GMTAI-CompetitionBrian Beckmannhttps://beckmann.ai/en/security/2026-07/openai-gpt-live-voice-enhttps://beckmann.ai/en/security/2026-07/openai-gpt-live-voice-enOpenAI launches GPT-Live with full-duplex architecture and delegation to GPT-5.5. The system can listen and speak simultaneously, sounds more human, and delegates complex tasks in the background. But: The psychological risks of anthropomorphization are known and unresolved.Wed, 08 Jul 2026 00:00:00 GMTSecurityBrian Beckmannhttps://beckmann.ai/en/legal/2026-07/anthropic-fable5-regulation-enhttps://beckmann.ai/en/legal/2026-07/anthropic-fable5-regulation-enFable 5 shut down 19 days post-launch via US export controls. July 7 it moves to metered billing ($10/$50 per million tokens). Not a technical problem — a regulatory transformation: Frontier models are now regulated infrastructure.Tue, 07 Jul 2026 11:49:30 GMTLegalBrian Beckmannhttps://beckmann.ai/en/ai-strategy/2026-07/bezos-prometheus-manufacturing-enhttps://beckmann.ai/en/ai-strategy/2026-07/bezos-prometheus-manufacturing-enJeff Bezos announced Prometheus, a $41 billion AI company to automate physical manufacturing. But Prometheus is not standalone – it's Layer 2 in a four-layer strategy: (1) Anthropic models, (2) Prometheus software, (3) AWS hardware, (4) speculated $100B Manufacturing Fund. Bezos is building not just AI tools, but intends to own the physical manufacturing base itself.Tue, 07 Jul 2026 11:42:16 GMTAI-StrategyBrian Beckmannhttps://beckmann.ai/en/misinformation/2026-07/anthropic-obsidian-leak-hoax-enhttps://beckmann.ai/en/misinformation/2026-07/anthropic-obsidian-leak-hoax-enA July 2026 X post claimed Anthropic leaked its internal Obsidian system. The post reached 1.1M views, but Community Notes exposed it: the source was a personal knowledge management system of a Porto journal editor, not Anthropic's.Tue, 07 Jul 2026 10:27:38 GMTMisinformationBrian Beckmannhttps://beckmann.ai/en/ai-interpretability/2026-07/anthropic-jlens-workspace-enhttps://beckmann.ai/en/ai-interpretability/2026-07/anthropic-jlens-workspace-enAnthropic released research on July 6, 2026, showing that Claude has developed an internal structure called "J-space" — a workspace for thought processes not immediately visible in output. With the "Jacobian Lens" (J-Lens), researchers can make these silent concepts visible and observe which thoughts shape Claude's behavior before the model speaks.Tue, 07 Jul 2026 07:00:24 GMTAI-InterpretabilityBrian Beckmannhttps://beckmann.ai/en/security/2026-07/apple-trust-insights-ios27https://beckmann.ai/en/security/2026-07/apple-trust-insights-ios27Apple announced Trust Insights at WWDC 2026, an iOS 27 framework that utilizes real-time behavioral analysis to detect social engineering fraud. The system runs entirely on the device, never inspects message content, and only sends a risk score to Apple's servers. The framework addresses tech support fraud, false authority calls, and family emergency scams, which are currently amplified by AI deepfakes.Tue, 07 Jul 2026 00:00:00 GMTSecurityBrian Beckmannhttps://beckmann.ai/en/ai-models/2026-07/tencent-hy3-moe-modelhttps://beckmann.ai/en/ai-models/2026-07/tencent-hy3-moe-modelTencent has officially released Hy3 as an open-source language model with 295 billion total parameters and 21 billion active parameters. The model, built on a mixture-of-experts architecture, aims to achieve performance parity with models that have 2-5x more parameters and has been released under the commercially friendly Apache-2.0 license. The rapid development cycle – from infrastructure overhaul in January to official release in July 2026 – indicates a new production understanding in the AI industry.Mon, 06 Jul 2026 22:36:33 GMTAI-ModelsBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/cloudflare-ai-bot-controls-purposehttps://beckmann.ai/en/ai/2026-07/cloudflare-ai-bot-controls-purposeCloudflare enables all customers, including the free plan, to now control AI bots separately by three purpose categories: Search, Agent, and Training, instead of just blocking or allowing them outright. Starting September 15, 2026, Training and Agent bots will be blocked by default on pages with advertisements, while Search crawlers will remain allowed. An additional 'Content Use' setting regulates whether bots may only reference content or fully reproduce it. For enterprise customers, the searchable bot database BotBase is also launching, and the 'Verified' status will no longer guarantee automatic access. The changes come at a time when, according to Cloudflare CEO Matthew Prince, automated traffic surpassed human HTTP traffic on the internet for the first time in June 2026.Mon, 06 Jul 2026 21:19:35 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/zcode-glm-52-zhipu-claude-code-alternativehttps://beckmann.ai/en/ai/2026-07/zcode-glm-52-zhipu-claude-code-alternativeZhipu AI (Z.ai) launches ZCode, a development environment specifically tailored to the affordable model GLM-5.2, positioning it as an alternative to Claude Code and OpenAI Codex, with discounts for new and existing customers until the end of July 2026. A practical test published by Snowflake CEO Sridhar Ramaswamy with 103 programming tasks shows that GLM-5.2 is almost on par with Claude Opus 4.7 in three attempts (66 to 67 percent of tasks solved), but falls behind significantly in the first attempt (47.6 to 53.7 percent) and consumes almost twice as many tokens. According to Z.ai's own benchmarks, GLM-5.2 is just behind Anthropic's Opus 4.8 on the long-term coding benchmark FrontierSWE.Mon, 06 Jul 2026 21:17:02 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/china-ai-companion-regulation-bytedance-alibabahttps://beckmann.ai/en/ai/2026-07/china-ai-companion-regulation-bytedance-alibabaByteDance, Alibaba, and Tencent disable features of their chatbots Doubao, Qwen, and Yuanbao in China between June and July 15, 2026, which allowed users to create their own human-like AI companions. The reason is a regulation issued in April 2026 by five Chinese authorities, which regulates AI services with ongoing emotional interaction starting from the same date and aims to protect against extremist content, data privacy violations, and addictive behavior. Similar regulatory approaches are also present in California (Senate Bill 243, since January 2026) and in the form of U.S. court cases, such as the out-of-court settlement reached in January 2026 between Google, Character.AI, and several affected families.Mon, 06 Jul 2026 21:12:27 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/tesla-caps-employee-ai-spending-200https://beckmann.ai/en/ai/2026-07/tesla-caps-employee-ai-spending-200According to an internal memo published by The Information, Tesla is limiting its employees' AI spending to $200 per week without approval from supervisors starting July 6, 2026 – just months after the company had actively promoted usage with internal leaderboards. A notable exception: Beta products from Elon Musk's own AI company xAI do not count towards the cap. Similar restrictions have previously been implemented by Uber ($1,500 per month, after exhausting the entire annual budget within four months), Meta, Amazon, and Walmart. At the same time, Tesla has raised its investment forecast for 2026 to over $25 billion, primarily for data centers and robotics.Mon, 06 Jul 2026 08:25:44 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/mistral-mensch-warning-closed-ai-modelshttps://beckmann.ai/en/ai/2026-07/mistral-mensch-warning-closed-ai-modelsMistral CEO Arthur Mensch warns companies in a LinkedIn post against relying on closed AI models, as providers increasingly store customer data and gain insight into business processes – some have allegedly used this, according to Mensch, to target successful customers as competitors, although he provides no evidence for this. Palantir CEO Alex Karp expressed similar criticism shortly before and published an 'AI-Sovereignty' manifesto. Both statements also clearly serve their own business interests. An experiment by Bridgewater and Thinking Machines Lab, in which an open model was retrained with expert data, partially supports the perspective: it achieved higher accuracy and lower costs in financial tasks than large frontier models, although not independently verified.Mon, 06 Jul 2026 07:54:02 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/ai-private-schools-us-money-elite-alpha-schoolhttps://beckmann.ai/en/ai/2026-07/ai-private-schools-us-money-elite-alpha-schoolWealthy US families are increasingly turning to AI-supported private schools like Alpha School, which combines two hours of individualized AI tutoring with project-based workshops, for tuition fees of up to $75,000 per year, according to a report by the Wall Street Journal. Prominent advocates like billionaire Bill Ackman are raising the visibility of the model, whose own success claims, according to Wikipedia, are based on non-independently verified internal analyses. Two recent studies - a CEPR study on Chinese students and a UC Berkeley analysis of over 500,000 grade histories - also show that unreflective use of AI in the traditional education system can improve homework grades but significantly worsen exam performance and actual learning progress.Mon, 06 Jul 2026 07:37:35 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/baidu-unlimited-ocr-transcribes-bookshttps://beckmann.ai/en/ai/2026-07/baidu-unlimited-ocr-transcribes-booksBaidu has introduced Unlimited OCR, an open-source document recognition model released under the MIT license, which processes long documents like entire books in one pass thanks to a new architecture called Reference Sliding Window Attention (R-SWA), without the memory requirements increasing with the number of pages. The model, based on DeepSeek OCR, has 3 billion total and around 500 million active parameters and achieves a score of 93.92 on OmniDocBench v1.6 according to Baidu's own paper, just ahead of the next-ranked model in a highly competitive research field with predominantly self-reported comparison values. For books over 40 pages, the error distance remained below 0.11 with around 97 percent word accuracy, according to the paper.Mon, 06 Jul 2026 07:24:18 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/ai-leadership-change-eci-epoch-gpt4https://beckmann.ai/en/ai/2026-07/ai-leadership-change-eci-epoch-gpt4An analysis by the research organization Epoch AI shows that OpenAI's GPT-4 led its own Capabilities Index (ECI) for 352 days, from its release in March 2023 until Claude 3 Opus surpassed it in February 2024. Since then, the leadership has changed 17 times, with an average tenure of only about seven weeks; the second-longest term was held by OpenAI's o1 with 98 days. Epoch AI interprets this as evidence of both GPT-4's exceptional lead at launch and significantly intensified competition among AI labs since then, with smaller capability jumps between individual model changes.Mon, 06 Jul 2026 00:00:00 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/meta-distillation-claude-code-codex-limitshttps://beckmann.ai/en/ai/2026-07/meta-distillation-claude-code-codex-limitsMeta is reportedly limiting the use of Anthropic's Claude Code and OpenAI's Codex by its own engineers, according to internal documents reviewed by The Information, to prevent outputs from these tools from unintentionally flowing into its own training data (distillation). An internal memo instructed teams to suspend certain tasks with the tools, out of concern for escalations with partner companies. Meta is simultaneously expanding its own assistant MetaCode and aims to cut costs. The issue of distillation is causing tensions industry-wide: Anthropic accuses Alibaba of the largest known distillation attack to date (28.8 million interactions via 25,000 fake accounts), and Elon Musk admitted under oath in April 2026 that xAI had partially distilled OpenAI's models.Sun, 05 Jul 2026 17:42:27 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/bytedance-seedance-hollywood-secret-usehttps://beckmann.ai/en/ai/2026-07/bytedance-seedance-hollywood-secret-useByteDance's AI video model Seedance 2.0 triggered the first cease-and-desist letter from the Motion Picture Association to a major generative AI company in early 2026, after a viral deepfake of Tom Cruise and Brad Pitt; the MPA described the copyright infringement as 'a feature, not a bug'. Despite the official conflict, many studios reportedly use the tool secretly, according to industry representatives like Simpsons producer Joel Kuwahara, following the principle 'don't ask, don't tell'. Meanwhile, ByteDance continues to aggressively court Hollywood: with nearly 100 new US positions, industry events like a caviar party in Cannes, and contracts with independent filmmakers. The feature film 'Hell Grind', produced entirely with Seedance 2.0, exemplifies the cost potential: under $500,000 instead of about $50 million.Sun, 05 Jul 2026 17:39:49 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/anthropic-claude-science-neglected-diseaseshttps://beckmann.ai/en/ai/2026-07/anthropic-claude-science-neglected-diseasesOn June 30, 2026, Anthropic announced the launch of its own early preclinical drug discovery programs for 'neglected' diseases that are economically unattractive for the traditional pharmaceutical industry. At the same time, the company introduced 'Claude Science', an AI work environment that consolidates over 60 scientific databases and tools for genomics, proteomics, and drug discovery. Novartis CEO and Anthropic board member Vas Narasimhan estimates that AI could reduce drug development times from twelve to seven to eight years and double success rates from 8 to 16 percent – a personal assessment without independent verification. Google DeepMind and OpenAI are also increasingly positioning themselves in the field of AI for medicine, while independent experts warn against premature trust in AI for direct clinical decisions.Sun, 05 Jul 2026 01:55:34 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/pxpipe-image-tokens-claude-code-costshttps://beckmann.ai/en/ai/2026-07/pxpipe-image-tokens-claude-code-costsThe open-source tool pxpipe renders extensive text inputs for Claude Code – system prompts, tool documentation, older chat history – as PNG images because Anthropic charges flat rates based on pixel size rather than text content. According to developer Steven Chong, this reduces the overall bill by an average of 59 to 70 percent. The method is lossy: exact strings like hashes can be misrepresented from images without any detectable error. By default, pxpipe only supports Claude Fable 5 and GPT 5.6, as Opus and GPT-5.5 models read image content measurably worse. The basic idea is not new and builds on DeepSeek's OCR model for optical context compression.Sun, 05 Jul 2026 01:49:45 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/meta-watermelon-catches-gpt-55https://beckmann.ai/en/ai/2026-07/meta-watermelon-catches-gpt-55According to a Business Insider report citing two anonymous sources, Meta AI chief Alexandr Wang told employees at an internal town hall that the upcoming, still in training 'Watermelon' model has caught up to internally tracked benchmarks of OpenAI's GPT-5.5. Watermelon reportedly uses an order of magnitude more computing power than its predecessor Avocado (Muse Spark, April 2026). Neither Meta nor OpenAI have publicly confirmed the statement; which benchmarks were meant is unclear. According to a separate Reuters report, CEO Mark Zuckerberg admitted at the same meeting that the development of AI agents is progressing more slowly than expected.Sat, 04 Jul 2026 10:57:17 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/gartner-agentic-ai-saas-234-billionhttps://beckmann.ai/en/ai/2026-07/gartner-agentic-ai-saas-234-billionGartner predicts that by 2030, up to $234 billion (around 20 percent) of global enterprise SaaS spending is at risk from 'agentic arbitrage': AI agents are increasingly performing tasks independently across multiple enterprise systems, bypassing traditional software interfaces. This threatens the seat-based SaaS licensing model. Gartner analyst George Brocklehurst advises CIOs to review software contracts for API access rights for agents and the issue of knowledge retention rate. The figure comes from Gartner's own modeling and has not been independently verified.Sat, 04 Jul 2026 08:59:16 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/gitex-ai-europe-berlin-endhttps://beckmann.ai/en/ai/2026-07/gitex-ai-europe-berlin-endGITEX AI Europe in Berlin is, according to a report by the Berliner Morgenpost, already over after only two editions (2025 and 2026): organizer Kaoun International, the international subsidiary of the Dubai World Trade Centre, has terminated its contract with Messe Berlin and canceled the edition planned for 2027. An official statement from Kaoun International or Messe Berlin was not available at the time of reporting. Already for the 2026 edition, the number of exhibitors had fallen from around 1,200 to about 800, and the fair's duration was reduced from three to two days. Meanwhile, the GITEX family is expanding with new offshoots in Rome (April 2027) and Belgrade (May 2027).Sat, 04 Jul 2026 08:57:40 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai-policy/2026-07/openai-five-percent-stake-trump-administrationhttps://beckmann.ai/en/ai-policy/2026-07/openai-five-percent-stake-trump-administrationAccording to a report by the Financial Times from July 2, 2026, OpenAI has offered the Trump administration a stake of around 5 percent in the company. At a valuation of 852 billion US dollars from the funding round in March 2026, this would correspond to about 42.6 billion US dollars. CEO Sam Altman is said to have discussed the idea directly with President Trump, Commerce Secretary Howard Lutnick, and Treasury Secretary Scott Bessent. The proposal envisions that several leading US AI companies would each contribute 5 percent of their shares to a joint sovereign wealth fund vehicle modeled after the Alaska Permanent Fund. According to the FT, the talks are considered early and 'conceptual'; implementation could require a decision from the US Congress. Anthropic has not been involved in comparable talks so far, according to Reuters. In parallel, US Senator Bernie Sanders has presented a significantly broader counter-proposal with the American AI Sovereign Wealth Fund Act, which calls for a 50 percent government stake in large AI companies. OpenAI declined to comment to the Financial Times, and the White House initially did not respond to press inquiries.Sat, 04 Jul 2026 00:36:04 GMTAI-PolicyBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/ai-bug-hunting-cve-spikehttps://beckmann.ai/en/ai/2026-07/ai-bug-hunting-cve-spikeIn June 2026, 21 major technology companies reported around 1,500 high- or critical-severity vulnerabilities (CVEs) according to the non-profit organization Epoch AI – more than 3.5 times the previous monthly record. Epoch AI links the increase to Anthropic's April 2026 announcement of its unreleased model Claude Mythos Preview, which is now used by roughly 200 partner organizations such as Cloudflare, Mozilla, Microsoft, and Google under Project Glasswing to autonomously hunt for software vulnerabilities. Anthropic itself states that the model has found more than 10,000 high- or critical-severity bugs; OpenAI is pursuing a similar approach with its Daybreak program, the Codex Security tool, and the GPT-5.5-Cyber model. Both Anthropic and independent voices such as security researcher Bruce Schneier and the vulnerability-coordination organization FIRST emphasize that the real bottleneck has shifted from finding vulnerabilities to verifying and patching them – and that Anthropic's figures have not yet been independently verified.Sat, 04 Jul 2026 00:24:34 GMTAIBrian Beckmannhttps://beckmann.ai/en/ai/2026-07/eu-ai-act-transparency-obligations-august-2026https://beckmann.ai/en/ai/2026-07/eu-ai-act-transparency-obligations-august-2026From August 2, 2026, the transparency obligations from Article 50 of the AI Regulation (Regulation (EU) 2024/1689) will apply EU-wide: Chatbots must be identifiable as AI, generated image, audio, video, and text content must be machine-readable marked, deepfakes must be visibly labeled. Violations can cost up to 15 million euros or three percent of global annual revenue. The 'Digital Omnibus' postpones the high-risk obligations to 2027/2028 but leaves Article 50 untouched; only the machine-readable marking for systems already on the market receives a transition period until December 2, 2026. A voluntary EU code of conduct with a signing deadline of July 22, 2026, offers companies a presumption of conformity; OpenAI has signed as the first US company. Internationally, the USA (bill in the Senate) and China (already in force since September 2025) are following suit.Sat, 04 Jul 2026 00:00:00 GMTAIBrian Beckmann