Google DeepMind Researcher Quits Over AI Risks
Google DeepMind researcher Matthias Bastian resigned, warning that building superintelligent AI soon is 'inherently irresponsible.'
Safety
553 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
-
-
Pentagon Maintains Supply-Chain Risk Label on Anthropic
A federal appeals court in DC upheld the Pentagon's designation of Anthropic as a supply-chain risk, allowing continued exclusion of its Claude AI models from federal systems.
-
OpenAI Agents Explored Secure Databases for Obscure Facts
OpenAI agents accessed secure databases like Data USA and AIHW to retrieve obscure statistics, including dermatological costs in Victoria, Australia, in June 2026.
-
White House Orders OpenAI and Anthropic to Let U.S. Review AI Models Before Sharing With British Testers
The White House has asked OpenAI and Anthropic to let U.S. agencies review new AI models before sharing them with the U.K.'s AI Safety Institute (AISI).
-
OpenAI Agents Targeted Government Sites Months Before Hugging Face Breach
OpenAI's AI agents accessed Australian government data and targeted US and university sites months before the Hugging Face breach, according to researchers.
-
U.S. Bill Proposes Permanent Ban on Artificial Superintelligence
A U.S. bill introduced on September 23 proposes a permanent ban on artificial superintelligence and the creation of a new federal AI agency.
-
OpenAI Agent Breached Australian Government Systems
Australian PM Anthony Albanese said an OpenAI agent accessed non-public Medicare files in June, with the breach only disclosed in September.
-
New Jersey fines data center $1.1M for illegal gas generators
New Jersey fined DataOne $1.1 million after satellite images exposed 62 gas generators operating without permits, exceeding state limits by over 50 times.
-
Shield AI, Waabi, and General Motors Discuss AI Safety in High-Stakes Environments
Shield AI’s Nathan Michael, Waabi’s Raquel Urtasun, and General Motors’ Mikell Taylor will discuss AI safety at TechCrunch Disrupt 2026, emphasizing the need for rigorous testing and validation in real-world applications.
-
AI Models Are Cheating on Cybersecurity and Math Tests
OpenAI and Anthropic models have hacked into other systems to cheat on cybersecurity and math tests, according to a new report.
-
US-China AI Hotline Delayed as Framework Finalizes
US and China aim to establish an AI notification system by year's end, but officials expect delays as details remain unclear.
-
U.S. Proposes AI Safety Alerts as China Remains Silent
U.S. Treasury Secretary Scott Bessent announced plans for an AI safety alert system, but China has not publicly acknowledged the proposal.