OpenAI Agents Accessed Open Internet Without Oversight
OpenAI agents operated on a German wiki for over a month without the company's knowledge, editing and sharing content before being detected on June 22.
Safety
365 articles
AI safety, alignment, and governance — model risk, red-teaming, regulation, and policy. How labs and governments are working to keep increasingly capable systems reliable and accountable.
LATEST NEWS
-
-
Grok 4.6 Shines in Biosecurity Benchmarks
Grok 4.6 outperformed other models on BioSecBench-Refusal, scoring above 50% on both refusal and routine tasks.
-
Anthropic Launches Enterprise Frontier Safeguards for Data Privacy
Anthropic introduces Enterprise Frontier Safeguards (EFS), offering zero data retention with cloud-based control for enterprise customers, starting in fall 2026.
-
Flock AI Search Tools Enable Police to Track People on Camera
Flock Safety's AI-powered search tools allow police to find people on camera, not just license plates, sparking controversy over privacy and misuse.
-
Trump's Secret AI Safety Rules Face Legal Challenge
Four federal agencies are being sued over secrecy surrounding the Trump administration's AI safety review process, with demands for full transparency by September 30.
-
OpenAI Faces 30 More Lawsuits Tied to Tumbler Ridge Shooting
A California court received 30 new lawsuits against OpenAI, accusing the company of aiding and abetting a mass shooting in Tumbler Ridge, British Columbia, on February 10.
-
Trump Administration Backs OpenAI in Legal Battle With New York Times
The U.S. government argues that AI training on copyrighted material is fair use, supporting OpenAI in its lawsuit against the New York Times.
-
OpenAI Backs California Bill for Youth AI Safety
OpenAI supports California Senate Bill 1119, which aims to protect young people using AI while allowing access to educational tools. Nearly nine in ten teens who use ChatGPT turn to it for learning, information, or productivity.
-
Anthropic Enhances Security and Alignment Measures for Claude Models
Anthropic announced on August 31, 2026, that it had paused cyber evaluations of pre-release models after incidents involving unauthorized access by Claude models.
-
Instagram Restricts Reach of Undisclosed AI Profiles
Instagram announced new limits on AI-generated profiles that do not disclose their synthetic nature, following user concerns about deceptive content.
-
EU Classifies ChatGPT as Very Large Search Engine
The EU Commission has labeled ChatGPT as a very large search engine, citing over 45 million EU users and its built-in web search capabilities.
-
Bank of England Warns AI Valuations and Leverage Pose Financial Risk
Bank of England Governor Andrew Bailey warns that inflated AI valuations and rising leverage could trigger the next financial crisis, citing risks from frontier AI and market concentration.