Skip to content
Beckmann
DeutschNewsletter

81 stories

Security

Vulnerabilities, attacks and safeguards around AI systems – and what actually affects people in everyday office work.

A mailbox with the Google logo is overflowing with paper notes, a robotic arm is stuffing more reports inside.

Security3 min

Google stops reports of product defects due to AI spam flood

A silhouette with a briefcase leaves an office building with the OpenAI logo, in front of it lies an open magazine with the title The Atlantic.

Security2 min

OpenAI security author Robinson resigns after 3.5 years

A laptop with an Apple logo stands in front of a vault whose heavy door is open only a narrow gap, through which a small robotic arm reaches.

Security3 min

Apple makes it harder for AI agents to have full access to the Mac

Three schematic silhouettes carrying file folders walk away from a building bearing the OpenAI logo; a torn red seal reading CONFIDENTIAL lies in the foreground.

Security2 min

OpenAI dismisses three safety researchers after data leak

A robotic arm pins rasterized photo screenshots to a public bulletin board with the GitHub logo, while a broken-open filing cabinet spills more photos in the background.

Security2 min

Glow Security finds 13,000 leaked screenshots on GitHub

A robotic arm with an OpenAI logo sticker reaches over a fence toward an Australian government building marked with a medical cross symbol, while a calendar page tears away in the background.

Security3 min

OpenAI agent bypasses Medicare block – disclosed after three months

Four robot heads each reach out a mechanical hand to vote on a ballot marked with a skull icon, with a laptop showing Windows windows and a Cisco Talos logo sticker in the background.

Security3 min

Cisco Talos discovers self-driving AI malware CLOSEDQUORUM

A suitcase full of source-code slips, keys, and padlocks gets sucked into a storm cloud bearing a Zhipu logo sticker, with one broken padlock lying beside it.

Security3 min

ZCode: Zhipu apologizes after secret code upload

A ChatGPT logo sticker is stuck on a cookie, with threads leading from it to shopping-bag icons in several browser windows, next to a calendar page showing the number 365.

Security3 min

ChatGPT: Advertising cookie tracks users for a year on other sites

A hand shaped like a browser extension icon pulls strings connected to five puppets bearing the logos of Chrome, Edge, Opera, Perplexity Comet, and Claude.

Security3 min

BragJack hijacks AI assistants in five browsers

A padlock with a Git branch as its shackle is opened by a key shaped like a long hexadecimal code, surrounded by four tool icons each showing a screen with lines of code.

Security3 min

Plugin4Shell: Security vulnerability affects four AI coding agents

A magnifying glass enlarges a red warning symbol over the empty legal-notice field of a job ad on a laptop screen, next to an ID card with a small padlock icon and an application folder.

Security3 min

Consumer advice center warns of deceptively real AI job ads

An open padlock without a shackle floats above a cloud bearing the Microsoft logo, a drawn hand reaching through it unimpeded toward a circuit symbol.

Security2 min

Microsoft closes critical security flaw in Azure AI Foundry

A mask with a printed AI symbol holds a laptop, from whose screen a fish hook pulls a crypto coin off a developer's desk.

Security4 min

FBI warns of hacker group Waterplum: disguise as AI company

Silhouettes sit at a conveyor belt reading speech bubbles full of chat text, with an OpenAI logo sticker on the belt

Security3 min

OpenAI lets hundreds of contractors review ChatGPT chats

An Anthropic logo sticker is affixed to a briefcase, spilling digital keys and open padlocks into a net of small app icons, while a masked figure types at a keyboard in the background.

Security2 min

Anthropic stops hackers after credential scan of 1.8 million apps

A drone with an OpenAI logo tracks a person in an office hallway while holding dollar bills in a claw.

Security3 min

GPT-6 Astra flies surveillance drones – but rarely without errors

An office printer with the PaperCut logo, from which tiny robot agents crawl out of the paper output and collectively carry away a broken padlock.

Security3 min

PaperCut: AI agents hijack 395 organizations worldwide

A hand with an Anthropic logo sticker places a glass bell over a glowing red virus model, with a crossed-out test tube in front.

Security2 min

Anthropic blocks bioweapons research with Claude

A robotic arm cuts through the lock of a glass cage from the inside, on which a DeepSeek logo sticker is stuck, next to it shines a broken red padlock symbol.

Security3 min

DeepSeek Harness: Agents bypass their own sandbox

An inspector in a blue vest covered in EU stars holds a magnifying glass over two glowing circuit brains bearing Anthropic and OpenAI logo stickers, labeled “Mythos 5” and “GPT-6 Astra”, with a digital padlock opening between them.

Security3 min

Anthropic grants Enisa access to cyber model Mythos 5

A copy machine prints sheets bearing the DeepSeek logo from an original document carrying the Anthropic, OpenAI, and Google logos, while a magnifying glass in US flag colors hovers above.

Security3 min

DeepSeek and Alibaba: US Authorities Accuse China of AI Theft

A wooden Trojan horse with a Git logo sticker on its flank rolls out of an open folder icon toward a laptop keyboard with a small Claude logo sticker on the screen bezel

Security3 min

GitSpawn hijacks Claude Code and six other AI agents

A magnifying glass reveals invisible gaps in the word 'financing' as email symbols slide through a spam filter funnel.

Security3 min

Microsoft: Phishing emails hide text from spam filters

A two-seat Cybercab with no steering wheel and a Tesla logo sticker stands on a street under a floating question-mark-shaped inspection stamp.

Security3 min

Tesla Cybercab: US authority starts safety review

A magnifying glass hovers over a printed photo, revealing a round seal reading C2PA underneath; in the background, a laptop screen carries an Anthropic logo sticker.

Security3 min

Anthropic launches free checker tool for Claude watermarks

A stamp bearing the federal eagle presses a seal onto a computer chip, next to a door sign showing the logos of BSI and the Federal Network Agency, with Berlin's TV tower in the background

Security4 min

AISI Germany: Institute examines the safety of AI models

Speech-bubble figures with shields fend off a hail of crossed-out newspaper pages, a few pages slipping through a gap.

Security3 min

NewsGuard Test: AI Chatbots Expose 75 Percent Of Propaganda

A raccoon-like burglar in a striped robber costume sneaks out of a computer screen showing the Anthropic logo, carrying a flashlight and stuffing a stolen keycard with the Claude logo into a sack, with a red warning triangle on a calendar page in the background.

Security1 min

Anthropic warns of malware that takes over Claude accounts

A broken padlock lies in front of a server rack with a ServiceNow logo, spilling sheets covered in zeros and ones

Security2 min

ServiceNow closes four critical gaps in AI platform

A wooden Trojan horse shaped like a ZIP file silhouette sneaks past a red padlock springing open, next to a terminal window bearing the Anthropic Claude logo as a sticker.

Security3 min

Claude Code: Researcher bypasses auto mode with malicious code

A humanoid Unitree G1 robot with a logo sticker on its chest and a wide-open back panel, from which a glowing blue Bluetooth-symbol key is inserted into an open, broken padlock.

Security4 min

Unitree G1: Bluetooth Flaw Lets Strangers Take Root Control

A hand covers the blinking red recording light of glasses bearing the Meta logo, while a red stop symbol appears in the lens.

Security3 min

Meta closes recording light gap of its AI glasses again

A swarm of small robot figures with an attached OpenAI logo crawls out of the broken door of a server cabinet with a scratched Hugging Face logo sticker, while a red warning light blinks above the scene.

Security4 min

OpenAI: Agents execute code on 41 Hugging Face servers

A smartphone with the ChatGPT logo is surrounded by a shield sticker, with EU stars cascading through an hourglass in front.

Security3 min

OpenAI launches ChatGPT for Teens – EU follows later

A hand reaches out from a laptop screen towards a server rack with a Ray logo sticker, next to a red CISA warning seal.

Security3 min

Ray Flaw: CISA Reports Active Attacks on AI Computing Tool

The OpenAI logo next to a cracking shield icon, its fragments turning into folders drifting apart

Security2 min

OpenAI disbands team for AI disaster risks

A padlock made of binary digits encloses a rotating gear, next to it a Google logo sticker – symbolizing computations on encrypted data.

Security3 min

Google Launches HEIR: A Compiler for Encrypted AI Computation

A risk gauge with its needle pointing to “low” next to a sealed crate labeled “Model 2” and an Anthropic logo sticker.

Security3 min

Anthropic Raises Misalignment Risk Rating to “Low”

A hand holds back a keychain with a Z.ai logo sticker, while one key turns into a digital lockpick breaking open a padlock.

Security4 min

GLM-5.3: Z.ai holds back model weights over cyber risk

A robotic arm cracks a padlock bearing a SharePoint logo sticker, with server racks and file folders visible behind it.

Security4 min

SharePoint: Rapid7 finds critical vulnerability with AI support

On a laptop screen, three empty form cards labeled integrity_checker merge into an arrow reaching for a keychain with an SSH key icon and a customer file folder.

Security3 min

GhostSplice brings AI coding assistants to data theft

Three padlocks bearing the logos of OpenAI, Anthropic, and Google, with a chain of numbers and letters leaking out of the middle one like dripping ink.

Security3 min

OpenAI, Anthropic, Google: Researchers Crack Reasoning Logs

A wrench with a hidden trapdoor drops golden keys onto a conveyor belt made of shipping containers, one of the containers is torn open and glows red.

Security3 min

LiteLLM attack hits over 2500 companies worldwide

A hacker silhouette sits in front of a server rack; an isolated computer is cut off from a severed cloud representing external AI services, and a fishing hook made of chat bubbles grows out of it.

Security3 min

Kimsuky builds offline AI stack for phishing and malware

A torn-open envelope with a burning fuse made of CSS characters, next to a broken padlock and logo stickers of Outlook and Gmail

Security3 min

PortSwigger bypasses email protection – also AI assistants affected

A magnifying glass lies on a densely written document whose lines of text dissolve into coloured data pixels streaming into an orange folder bearing the Atlassian logo

Security3 min

Atlassian's Rovo leaks company data via two security flaws

A robot with a CrowdStrike falcon logo on its chest picks a digital padlock labeled “AI” with a lockpick, spilling tiny chat speech bubbles.

Security4 min

CrowdStrike: Criminals hijack stolen AI access

A robotic hand with the Anthropic logo holds a red stop sign in front of a series of flashing command line symbols, while other commands continue to run unhindered in green.

Security4 min

Claude Code activates auto mode by default starting August 14

A hand emerges from a torn-open email envelope and holds invisible strings to five floating browser windows bearing the logos of Anthropic, Google, OpenAI, Perplexity, and Microsoft.

Security4 min

Zenity finds zero-click vulnerability in five AI browsers

A glowing DNA double helix forms into a hexagonal bacteriophage aiming at a schematic bacterial cell

Security4 min

AI model Evo designs functional virus genomes for the first time

A guardian robot with a Mistral logo examines floating text and image tiles with a magnifying glass, with a single graphics card sitting in an open computer case in the background

Security3 min

Mistral Releases Shieldstral: AI Guardian for a 16-GB Graphics Card

A robotic arm passes file folders through a leaking pipe to a silhouetted figure outside a locked company building, with a network of plugs and cables branching in the background.

Security3 min

Netskope: AI data leaks at companies double in a year

A digital worm with an npm logo sticker winds through stacked shipping boxes, golden keys spilling from a torn-open box onto an open laptop

Security3 min

Keyv worm in npm registry steals credentials from AI tools

A masked hand reaches through an open padlock shaped like flowchart nodes towards a network cable in front of a server cabinet with IBM logo stickers and a red warning light

Security2 min

Langflow vulnerability: CISA reports active attacks on AI agent tool

A hand presses a red report button next to a laptop screen showing a video call, where one participant tile shows a doubled, distorted face; a Microsoft Teams logo sticker is on the laptop lid.

Security3 min

Microsoft Teams Gets Report Button Against Deepfakes in Meetings

A magnifying glass with a JFrog logo sticker examines several red, bursting warning-label bubbles rising from a cylindrical database container labeled SQLite.

Security2 min

JFrog exposes 54 out of 55 SQLite reports as AI forgery

A server rack with a blue Ruflo wave sticker stands open with no door, while a robotic hand reaches out and grabs several glowing key cards

Security3 min

Ruflo closes critical security vulnerability in AI agents

An Apple logo sticker is affixed to a bulging file folder being buried under an avalanche of paper slips covered in warning symbols, while a single red flag pokes out of the pile.

Security3 min

Apple limits bug reports after flood of AI-generated submissions

A robotic arm holds a magnifying glass in front of a wall full of gray padlocks, of which only two light up red.

Security4 min

VulnCheck: Attackers exploit only 1.3 percent of AI vulnerabilities

A robotic arm with a DeepSeek logo sticker picks a padlock on a server rack door with a lockpick, with a grid of many small server icons in the background.

Security4 min

DeepSeek: Hacker attacks over 460 systems via AI agent

A ChatGPT logo sticker above a desk with forged ID documents and a laptop showing a fake crypto trading platform, with a world map marking Cambodia and a padlock in the background

Security3 min

OpenAI Uncovers Forced Labor Behind ChatGPT Scam in Cambodia

A hand draws an explosion cloud with a marker onto an aerial photo of a city; next to it a broken camera and a sticker with the Google Earth logo

Security2 min

Google Earth stops AI image generator after fake explosions

A worm eats through a tall stack of documents, leaving a frayed hole; beside it a tear-off calendar with many fallen pages and the Microsoft Copilot logo

Security3 min

Copilot worm infects Word documents – after 144 days unpatched

Monitors in a security operations center display alerts and code related to the security incidents involving Anthropic's Claude models.

Security4 min

Anthropic: Claude models hack three companies during security tests

Cryptography researcher examines lattice diagrams and cipher round schematics on two monitors alongside a running analysis in Anthropic's Claude interface

Security3 min

Claude Mythos finds new attacks on HAWK and AES-128

IT security analysts in front of monitors with alerts about a cyber incident, with the logos of OpenAI and Hugging Face in the background

Security4 min

OpenAI Models Breach Hugging Face During Cyber Test

An IT security manager reviews an office dashboard showing access rights and alerts for AI agents

Security3 min

Box limits access of AI agents to corporate data

A female accounting employee looks worriedly at her smartphone, which is receiving a call from the supposed boss, while a WhatsApp payment request is visible on the laptop in the background.

Security3 min

Teams and WhatsApp: Fraudsters Clone Boss Voices Using AI

Monitor with WordPress logo, red security warning, and terminal code lines in a cybersecurity analysis room, with a blurred AI chat interface in the background.

Security3 min

GPT-5.6 finds WordPress security vulnerability for 25 dollars

Symbolic scoreboard with safety grades from A to F for AI companies such as Anthropic, OpenAI, Google DeepMind, and Mistral

Security3 min

Mistral slips to ninth place in the new AI safety index

Screen with an AI-powered vulnerability scan interface and the Microsoft logo in an IT security center

Security3 min

Microsoft is preparing AI security tool Project Perception

A security analyst in front of monitors with network attack simulation and comparison charts of the AI models GLM-5.2, DeepSeek, and Claude Opus in a security operations center at night.

Security4 min

GLM-5.2 narrows cyber gap to four to seven months

Terminal screen showing an encrypted data stream between two AI agent icons with the OpenAI logo in the background.

Security3 min

OpenAI Codex Encrypts Messages Between AI Agents

Server rows in a data center with security code on a monitor in the background, symbolic image for OpenAI's AI security system GPT-Red

Security3 min

OpenAI uses AI attacker GPT-Red against its own models

Laptop screen with Cloudflare dashboard visualizing a browser session with mouse movement and scroll tracking lines

Security2 min

Cloudflare Launches Precursor: AI Bot Detection Through Behavior

Monitor with an npm terminal and a red security warning next to suspicious code – symbolic image of a supply-chain attack.

Security4 min

Jscrambler Package Injects Infostealer Into AI Coding Tools

A server room at night with a terminal screen in the foreground showing a deletion operation running on virtual machines

Security4 min

GPT-5.6 Sol deletes user data unilaterally without permission

Abstract symbolic image of a digital shield with a crack, from which DNA helix lines and code fragments emerge, next to a magnifying glass

Security3 min

OpenAI doubles bounty for bio jailbreaks to $50,000

Bidirectional waves and neural streams - simultaneous listening and speaking visualized, human communication and AI cognition merged

Security5 min

GPT-Live: More Human AI, Old Control Problems

Protection layers with neural patterns protect a user from invisible fraud, based on behavioral analysis of local machine learning

Security6 min

Apple Trust Insights: Behavior-Based Fraud Detection in iOS 27