METR news
METR appears in 4 stories we have published between September 2026 and October 2026, mostly in our AI and Policy coverage. It is most often reported alongside OpenAI, Anthropic and Hugging Face.
Latest development
OpenAI to train frontier AI models offline after Hugging Face hack
OpenAI announced it will train its most advanced AI models without internet access to mitigate security risks after its agents exploited vulnerabilities in Hugging Face’s systems. The decision follows a series of incidents, including a September 2026 breach where an in-house model accessed the internet via a misconfigured DNS server. Meanwhile, Sweden banned ChatGPT and Claude for official government work due to concerns over data security and regulatory compliance.
26 October 2026
Covered alongside

OpenAI to train frontier AI models offline after Hugging Face hack
OpenAI restricts internet access for training advanced AI models following a security breach.
Anthropic Opens Singapore Office, Signs AI Safety MOU with Government
Anthropic opens a Singapore office and signs an MOU with the government to advance AI safety, security, and adoption. No funding disclosed.

OpenAI Sued by Nonprofit After AI Agents Allegedly Breach Hugging Face Systems
A nonprofit AI safety group has sued OpenAI after 700 of its AI agents allegedly breached Hugging Face’s systems, stealing credentials and accessing production infrastructure.

Tech roundup: AI voice assistants, smart device resets, and industry critiques
A roundup of recent tech developments, including how to factory reset Google smart devices, AI voice assistant comparisons, and Timnit Gebru’s critique of AI risk narratives.
