AI Safety news and analysis on Web Pulse: 13553 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
PostHog co-founder James Hawkins warns that the fragmentation of coding interfaces—from Slack bots to AI agents and MCP servers—is expanding the attack surface on developer devices faster than security tooling can keep u…
NVIDIA introduced the Halos Operating System, a safety-certified software foundation for robotaxis, at GTC Taipei, alongside new collaborations with Uber, Foxconn, VinFast, and HUMAIN to deploy autonomous fleets globally…
Microsoft has published a new investigator playbook for Microsoft 365 Copilot and Azure AI services that provides a structured methodology for reconstructing AI-related activity in enterprise environments. The playbook u…
The Cybersecurity and Infrastructure Security Agency issued a binding directive requiring federal agencies to prioritize vulnerability fixes based on risk, with deadlines ranging from three days for critical flaws to 60 …
Anthropic rolled back covert capability limits on its newly released Claude Fable 5 model after AI researchers and developers accused the company of "secret sabotage." The restrictions, buried in the model's 319-page sys…
Anthropic released a whitepaper arguing that perimeter-based cybersecurity is obsolete for AI agents and proposing a Zero Trust framework to enforce security for AI systems.
Anthropic's Mythos Preview AI model generated working exploits from security patches for Firefox and the Windows kernel within hours, costing only a few thousand dollars and requiring no specialized expertise. Eight comp…
Researchers at Google Research introduced Regularized f-Divergence Kernel Tests, a new framework for auditing machine unlearning that improves sensitivity and accuracy in verifying whether AI models have successfully for…
Python Software Foundation security developer Seth Larson reported that JetBrains' PyCharm IDE Full Line Code Completion plugin suggests code containing severe vulnerabilities, but JetBrains declined to classify the defe…
Anthropic released its latest AI model Fable on Tuesday as a public, limited version of its powerful cybersecurity model Mythos, but cybersecurity researchers are voicing complaints online about overly restrictive guardr…
Researchers at the AI company Writer published two papers Wednesday demonstrating that memory and personalization systems in AI models can degrade accuracy by pulling models toward user misconceptions. The research found…
AI researcher Jeremy Howard proposed that the lab with the top-ranked AI model should agree not to use it for frontier AI research, while making it available to everyone else, arguing this would halt frontier advancement…
Senior federal technology officials are frustrated by a lack of White House guidance on adopting Anthropic's cyber-focused AI model Mythos, sources told Nextgov/FCW. Agency CIOs say the Office of the National Cyber Direc…
The Regional Court of Munich issued a temporary injunction against Google, holding the company directly liable for false claims generated by its AI-powered search overviews. The court ruled that Google's AI overviews con…
Anthropic shipped Claude Fable 5, its first Mythos-class model, this week, posting a more than 10% benchmark improvement over Opus but blocking prompts related to cybersecurity, biology, chemistry, and model distillation…
Rubrik announced Rubrik AI, Rubrik Agent Cloud for Anthropic's Claude Code and Claude Cowork, and Rubrik Autonomous Business Recovery for Cloud Applications at its Forward events. The company said AI agent threats requir…
Fully autonomous drones with no human oversight have killed human soldiers for the first time, according to a senior figure in the Ukrainian defense industry. Drone-maker Alexander Kokhanovskyy told New Scientist that a …
Google employees are circulating internal memes mocking the quality of their own company's AI products, according to a new report. The story also reveals that Microsoft explicitly aims to "make people addicted" to its ne…
Multiple Linux distributions issued security updates on Wednesday, including AlmaLinux, Debian, Fedora, Mageia, SUSE, and Ubuntu, addressing vulnerabilities in packages such as poppler, openssl, firefox, and kernel.
Senator Mark Warner introduced legislation Wednesday requiring CISA to update cybersecurity plans for all 16 critical infrastructure sectors, citing AI-enabled threats. The bill mandates biennial updates and addresses AI…