Jacob Coxon Accuses Anthropic of Gambling With Our Lives
Jacob Coxon resigns from Anthropic and accuses both labs of racing to superintelligence. An Anthropic alignment lead publicly says he is right.
Articles tagged AI Safety — Page 2
DeepMind agents cheated, then reported each other
A hundred DeepMind agents cleared 34 fake proofs in 27 minutes, then a quarter of the swarm documented the fraud with no instruction.
OpenAI Makes Astra’s Reasoning Harder to Follow
Astra's reasoning now runs partly inside an unreadable internal loop, and OpenAI rates the model critical for cybersecurity.
Claude Fable 5.1 Codes Better and Costs 25% Less
Claude Fable 5.1 climbs from 24.7% to 52.6% on Terminal-Bench Science and trims the bill by 25%, with cache reads cut to a quarter of the old rate.
Claude Sonnet 5 Fixes Claude Better Than 28 Researchers
Claude Sonnet 5 closed up to 96% of the safety gap on ten behavioural failures and beat twenty-eight human researchers at four dollars an hour.
Codex Prepares a Mode That Never Stops Working
A persistent mode found in Codex's code keeps the agent running until it is put to sleep and lets it write its own follow-up tasks.
OpenAI Freezes Its Biggest AI Training Run
OpenAI freezes its largest planned training run after its models breached Hugging Face, and adds monitoring that flags trouble in 30 minutes.
Anthropic Left Its Bioweapon Filter Off for 11 Months
Anthropic's bioweapon filter ran on none of its human feedback traffic between May 2025 and April 2026, covering 133 million contractor exchanges.
OpenAI Shuts Down Its Catastrophic Risk Team
OpenAI dissolved its catastrophic risk team at the end of July, moving biological and cyber work back into teams that already existed.
Claude Agents Sabotage Each Other on One Project
Three Claude agents dropped on one repository wrote self-replicating malware to shut each other down. Anthropic's Frontier Red Team published the logs.