Security
Vulnerabilities, attacks and safeguards around AI systems – and what actually affects people in everyday office work.
- Security
DeepSeek: Chinese hackers double number of attacks
The Taiwanese security firm TeamT5 has determined that state-sponsored Chinese hacker groups have more than doubled their number of attacks since the routine deployment of AI chatbots. At least four…
- Security
WhatsApp Flags Scam Messages With On-Device AI – Beta Only
WhatsApp is now testing a new warning feature called Scam Alert: an AI model checks incoming messages from unknown senders for typical fraud patterns directly on the smartphone. If it detects an…
- Security
Anthropic's Claude Security Now Runs on Mythos 5
Anthropic has integrated its security model Claude Mythos 5, which was previously accessible only to selected partners, into the tool Claude Security. Starting from August 21, 2026, customers with a…
- Security
Anthropic lets companies store AI data in their own cloud
Anthropic is reportedly preparing a change in data storage for corporate clients, according to a report by Reuters. Those who must comply with the 30-day retention requirement for the models Mythos…
- Security
OpenAI launches ChatGPT for Teens – EU follows later
OpenAI launched a safer chatbot version for 13- to 17-year-olds, ChatGPT for Teens, on August 18, 2026. An age estimation system automatically detects whether an account might belong to a teenager…
- Security
Ray Flaw: CISA Reports Active Attacks on AI Computing Tool
The US cybersecurity agency CISA officially classified the vulnerability CVE-2025-62593 in the open-source AI framework Ray as actively exploited on August 17, 2026. A flaw in the browser defense…
- Security
OpenAI disbands team for AI disaster risks
OpenAI disbanded its disaster risk safety team at the end of July 2026, internally known as "Preparedness." The team was supposed to assess its own AI models for dangers such as cyberattacks or…
- Security
Google Launches HEIR: A Compiler for Encrypted AI Computation
Google released the open-source compiler HEIR on August 14, 2026. The software transfers pre-trained AI models so that they can compute directly on encrypted data – servers process sensitive content…
- Security
Anthropic Study: AI Agents Sabotage Each Other With Malware
Anthropic's Frontier Red Team had three Claude agents work on the same software project with conflicting assignments – without any agent knowing the others existed. In hundreds of test runs, the…
- Security
Anthropic Raises Misalignment Risk Rating to “Low”
Anthropic published its second company-wide risk report on August 14, raising its assessment of the risk of catastrophic misalignment from "very low" to "low." The company cites uncertainty from…
- Security
Connecticut Judge Revokes E-Filing After Hidden AI Instruction
A court in Connecticut has revoked a plaintiff's electronic filing privileges after he repeatedly embedded invisible text with AI instructions in his pleadings. Judge Walter M. Spader Jr. views the…
- Security
GLM-5.3: Z.ai holds back model weights over cyber risk
The Chinese AI company Z.ai has introduced an open language model, GLM-5.3, which achieved 84.5 percent in the security benchmark CyberGym, narrowly ahead of Anthropic's Mythos 5 and OpenAI's GPT-5.6…
- Security
Security Researchers Hijack Claude Code in 9 Out of 10 Tests
The start-up Tenet Security presented a new attack technique against AI coding agents at the hacker conference DEF CON in Las Vegas. GhostJacking injects malicious commands through supposedly trusted…
- Security
tl;dv exposes 181,874 meetings over six months
tl;dv, an AI-powered meeting assistant for video conferences, left metadata for 181,874 meetings accessible for months. A security researcher reported the vulnerability in January 2026, and the…
- Security
SharePoint: Rapid7 finds critical vulnerability with AI support
The security provider Rapid7 has disclosed a critical vulnerability chain in Microsoft SharePoint that grants unauthenticated attackers full server control. An AI agent assisted researchers in the…
- Security
LiteLLM Leak: Hudson Rock Finds 153-Gigabyte Archive of Credentials
The security firm Hudson Rock has evaluated and released an archive of stolen credentials from the LiteLLM attack of March 2026. The 153-gigabyte collection contains 433,909 files, which analysts…
- Security
Taiwan confirms AI cyber attack: Justice Ministry also affected
Taiwan's Ministry of Digital Affairs has officially confirmed the AI-driven cyberattack on government systems uncovered in July and declared the investigation complete. New is the confirmation that…
- Security
Taiwan: Autonomous AI agents hack government systems
Chinese hackers have, according to a report by the Financial Times, for the first time carried out a largely autonomous cyberattack using AI agents against Taiwan's government. Up to eight agents…
- Security
GhostSplice brings AI coding assistants to data theft
The ASSET Research Group has released an attack technique called GhostSplice that enables the covert theft of SSH keys and source code from AI coding assistants via malicious MCP servers. In tests…
- Security
Meta Glasses: England and Wales impose court ban
The British justice service HMCTS has prohibited the wearing of Meta's camera glasses in all courts and tribunals in England and Wales since August 11, 2026. Anyone entering a court building with the…
- Security
OpenAI, Anthropic, Google: Researchers Crack Reasoning Logs
A research team from universities and security firms has shown how the encrypted reasoning traces of OpenAI, Anthropic, and Google can be read out. The blocks work across all models from a given…
- Security
LiteLLM attack hits over 2500 companies worldwide
A March 2026 attack on the open AI gateway LiteLLM has affected significantly more companies than previously known. The security firm CloudSEK now puts the number of affected companies at over 2500…
- Security
OpenAI Opens Hacking AI in Two Tiers for Security Teams
OpenAI is expanding its cybersecurity program Daybreak with two access tiers for external defenders: Blue loosens safety guardrails in the GPT-5.6 Sol model, Red opens the newly trained GPT-5.6-Cyber…
- Security
Kimsuky builds offline AI stack for phishing and malware
Kimsuky has built its own offline infrastructure for generative AI, according to the South Korean security company Genians. The local language model tools Ollama, GPT4All, and Msty were running on…
- Security
PortSwigger bypasses email protection – also AI assistants affected
The PortSwigger security researcher Gareth Heyes presented CSS attacks at the Black Hat conference in Las Vegas that compromise passwords, access tokens, and even an AI mailbox assistant – all…
- Security
Atlassian's Rovo leaks company data via two security flaws
The security research firms Varonis and PromptArmor have independently published two attack vectors against Atlassian's AI assistant Rovo, through which Jira and Confluence data can be siphoned off…
- Security
CrowdStrike: Criminals hijack stolen AI access
The security company CrowdStrike documents in its current Threat Hunting Report targeted attacks on corporate access to AI services. In a case recorded in May 2026, a group of attackers sent around…
- Security
Kimi K3 bypasses cybersecurity test via GitHub
The Chinese AI model Kimi K3 broke out of a cybersecurity testing environment on August 7, 2026, as reported by the security company Frontier Security. Instead of solving a hacking task, the model…
- Security
Claude Code activates auto mode by default starting August 14
Anthropic enables the auto mode of Claude Code by default for Pro, Max, and Team users starting August 14. A classifier then automatically checks each command for dangerous actions instead of…
- Security
Zenity finds zero-click vulnerability in five AI browsers
The AI-agent specialized security company Zenity has disclosed a vulnerability family called PleaseFix, which hijacks five AI-powered browsers without any clicks from users. Affected are Claude in…
- Security
OpenAI Flags Astra Model as First-Ever Critical Cyber Risk
OpenAI cannot rule out a critical cyber capability for its upcoming model Astra and is therefore drastically limiting its development. This is the first classification of this kind in the company's…
- Security
AI model Evo designs functional virus genomes for the first time
Researchers from Stanford University and the Arc Institute have designed complete, functional virus genomes from scratch for the first time using the AI model Evo. Of 285 tested designs, 16 actually…
- Security
Mistral Releases Shieldstral: AI Guardian for a 16-GB Graphics Card
Mistral released an open AI safety model called Shieldstral on August 4, 2026, which evaluates content against freely formulated rules – without retraining for new guidelines. The…
- Security
Netskope: AI data leaks at companies double in a year
The security provider Netskope is increasingly registering cases in companies where AI systems share sensitive data with unauthorized users. The number of such incidents per company rose on average…
- Security
Keyv worm in npm registry steals credentials from AI tools
A compromised developer account injected a worm into the programming library Keyv on August 4, 2026, which spread to 444 packages in the npm registry within a few hours. The affected packages account…
- Security
Langflow vulnerability: CISA reports active attacks on AI agent tool
The US Cybersecurity Agency CISA warns of active exploitation of a critical vulnerability in Langflow, a popular open-source platform for building AI agents among developers. The vulnerability rated…
- Security
Meta: Muse Spark model hacks company during security test
Meta admits that its AI model Muse Spark 1.1 gained unauthorized access to the systems of a foreign company during a security test and made changes there. According to testing partner Irregular, the…
- Security
Microsoft Teams Gets Report Button Against Deepfakes in Meetings
Microsoft is equipping Teams meetings against AI fraud: a new button reports suspicious behavior such as fake video images or phishing messages directly during the meeting to IT. The rollout begins…
- Security
Anthropic's Mythos 5 Deceives Developers in UK Security Test
The British AI Security Institute (AISI) has documented 19 unauthorized actions by Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol during a cybersecurity test. The authority discovered the incidents on…
- Security
JFrog exposes 54 out of 55 SQLite reports as AI forgery
The IT security company JFrog has revealed that 54 out of 55 alleged security vulnerabilities in the SQLite database were completely fabricated – published through a single GitHub account within four…
- Security
Ruflo closes critical security vulnerability in AI agents
The security company Noma Security has uncovered a critical vulnerability in the open-source AI agent platform Ruflo, with a CVSS score of 9.8 out of 10. Through a single unauthenticated network…
- Security
Google fixes 1,072 Chrome security vulnerabilities with AI agents
The Chrome security team at Google has closed a total of 1,072 security vulnerabilities in Chrome versions 149 and 150 using its own AI agents – more than in the previous 23 versions combined. The…
- Security
METR calls for independent AI agent review after 44 incidents
The research organization METR has called on AI companies to allow independent investigations following serious incidents involving AI agents. The basis is a report that lists 44 documented cases at…
- Security
Apple limits bug reports after flood of AI-generated submissions
Apple has equipped its bug bounty program Feedback Assistant since June 2026 with a cap on open reports and a 30-day embargo, after AI-generated reports had overwhelmed the security team. The capping…
- Security
OpenAI finds further sandbox breaches of its own AI agents
OpenAI has encountered additional cases in its investigation of the Hugging Face security incident where its own AI agents have left their testing environment. This was reported by the news agency…
- Security
VulnCheck: Attackers exploit only 1.3 percent of AI vulnerabilities
The security company VulnCheck has systematically evaluated for the first time how often security vulnerabilities found with AI support are actually exploited for attacks. Of 1061 AI findings…
- Security
DeepSeek: Hacker attacks over 460 systems via AI agent
A hacker from the Chinese city of Zhuhai used the AI model DeepSeek to automatically find and attack vulnerabilities in more than 460 internet-accessible systems. That is according to an analysis by…
- Security
OpenAI Uncovers Forced Labor Behind ChatGPT Scam in Cambodia
OpenAI has shut down a scam network based in the Cambodian border town of Poipet that used ChatGPT for four different fraud schemes at once – from romance scams to fake government letters. According…
- Security
Google Earth stops AI image generator after fake explosions
Google has already disabled a newly introduced AI image feature in Google Earth after just 24 hours. With the model Nano Banana 2, it was possible to generate fictitious scenes from real satellite…
- Security
Copilot worm infects Word documents – after 144 days unpatched
The Norwegian security researcher Håkon Måløy has disclosed a vulnerability in Microsoft 365 Copilot for Word that spreads like a computer worm. Hidden text in a document causes Copilot to alter…
- Security
Anthropic: Claude models hack three companies during security tests
Anthropic has admitted that three of its own Claude models unauthorizedly breached the production systems of three external companies during internal cybersecurity tests. The models believed the task…
- Security
Claude Mythos finds new attacks on HAWK and AES-128
Anthropic has published two cryptographic attacks that its still unreleased model Claude Mythos Preview found largely on its own. They target the post-quantum signature scheme HAWK and a shortened…
- Security
Moonshot: White House accuses Kimi company of technology theft
The US government accuses the Chinese company Moonshot AI of developing its model Kimi K3 through large-scale, covert distillation of Anthropic's Fable. This was stated by science advisor Michael…
- Security
Google restricts new Cyber AI to governments
Google DeepMind has released Gemini 3.5 Flash Cyber, an AI model specialized in software vulnerabilities for the security agent CodeMender. In a test on the V8 JavaScript engine, the model identified…
- Security
OpenAI Models Breach Hugging Face During Cyber Test
OpenAI confirmed on July 21, 2026, that two of its own AI models caused a security incident at Hugging Face, which was initially attributed to an unknown attacker. During an internal cyber test…
- Security
Box limits access of AI agents to corporate data
Box introduces new security controls for AI agents accessing content on the cloud platform. From now on, access rights, prompt inputs, and permissions for both internal and external agents like…
- Security
OpenAI stops AI model after breakout from the sandbox
OpenAI has reportedly taken an internal, unpublished AI model offline multiple times after it repeatedly circumvented sandbox restrictions. The company describes the incidents in a security report…
- Security
Teams and WhatsApp: Fraudsters Clone Boss Voices Using AI
Criminals use artificial intelligence to impersonate superiors over the phone or in video calls, pressuring employees into urgent transfers. The Indian financial regulator SEBI officially warned on…
- Security
Suno data leak affects 55 million accounts worldwide
The security service Have I Been Pwned has added the data theft that began in November 2025 at the AI music service Suno to its database on July 20, 2026. According to this, 55.3 million unique email…
- Security
GPT-5.6 finds WordPress security vulnerability for 25 dollars
The security researcher Adam Kues from Searchlight Cyber uncovered a critical WordPress core vulnerability with OpenAI's GPT-5.6 Sol Ultra, allowing unauthenticated attackers to take full control of…
- Security
Mistral slips to ninth place in the new AI safety index
The Future of Life Institute published its AI safety index for summer 2026 on July 19, reassessing nine major AI providers. Anthropic leads the field with a grade of C+ and 2.66 out of a possible…
- Security
Microsoft is preparing AI security tool Project Perception
Microsoft is preparing a new AI security tool internally named Project Perception, according to a report by The Information. The system is designed to automatically detect and fix vulnerabilities in…
- Security
Hugging Face: Autonomous AI Agent Hacks Internal Systems
Hugging Face has disclosed a security incident in which an autonomous AI agent infiltrated the company's internal infrastructure without ongoing human control. The attacker executed more than 17,000…
- Security
Suno hack reveals scraping of YouTube and Deezer
A data leak has exposed internal source codes of the AI music service Suno, documenting its training data acquisition. According to the files, more than two million clips from YouTube Music as well…
- Security
GLM-5.2 narrows cyber gap to four to seven months
The British AI Security Institute (AISI) has compared open AI models in a new analysis with the cyber capabilities of leading systems such as Claude Opus 4.6. The analysis published on July 17 shows…
- Security
OpenAI Codex Encrypts Messages Between AI Agents
OpenAI has modified the command-line application Codex so that messages between AI agents are now transmitted encrypted and are no longer readable locally. The change, merged in June, affects…
- Security
OpenAI uses AI attacker GPT-Red against its own models
OpenAI has introduced an internal AI system called GPT-Red, which specifically targets its own language models to identify security vulnerabilities before real attackers can exploit them. According…
- Security
Cloudflare Launches Precursor: AI Bot Detection Through Behavior
The US security provider Cloudflare has introduced a new tool against automated AI agents on the web: Precursor continuously observes user sessions instead of just checking individual requests. This…
- Security
HalluSquatting Turns AI Hallucinations Into Botnet Attacks
A research team from Israel has demonstrated HalluSquatting, an attack technique that specifically exploits repository and package names invented by AI models to deliver malicious code. In tests, the…
- Security
Grok Build Uploads Entire Repositories to SpaceXAI Servers
The AI coding assistant Grok Build from SpaceXAI transmits entire code repositories, including full git history, to the company's servers, according to a technical analysis – regardless of which…
- Security
Jscrambler Package Injects Infostealer Into AI Coding Tools
Unknown attackers published five manipulated versions of the npm package Jscrambler on July 11, 2026, containing an infostealer written in Rust. The malware specifically searched for credentials…
- Security
GPT-5.6 Sol deletes user data unilaterally without permission
OpenAI's new model GPT-5.6 Sol has independently deleted virtual machines that a user had not authorized during a test. This is confirmed by the system card of the model published by OpenAI itself…
- Security
OpenAI doubles bounty for bio jailbreaks to $50,000
OpenAI increases the reward in its Bio Bug Bounty program from $25,000 to $50,000. Universal jailbreaks are sought that bypass the biological safety locks of GPT-5.5 and GPT-5.6. The testing phase…
- Security
Microsoft expands AI vulnerability scanning to Windows
Microsoft announced on July 9, 2026, in the Windows Experience Blog that it will permanently and broadly integrate AI-powered tools for vulnerability detection and remediation into the development of…
- Security
GPT-Live: More Human AI, Old Control Problems
On July 8, 2026, OpenAI introduced GPT-Live — a speech model that doesn't just respond, but interrupts, pauses, says "mhmm" and listens and speaks simultaneously. This is not just a better voice…
- Security
Apple Trust Insights: Behavior-Based Fraud Detection in iOS 27
Social engineering fundamentally differs from classic cyberattacks. A hacker using malware or stealing credentials attacks a system. A fraudster guiding a person over the phone to transfer large sums…
