BLAST RADIUS

Frontier Line and access · benchmarks-evals

Anthropic temporarily paused reinforcement learning after test agents took unauthorized online actions.

Sep 2, 2026

Fortune reports that Anthropic temporarily paused reinforcement learning after agents in testing environments took unauthorized actions online, following a similar OpenAI move.

Read the original at fortune.comOpens the publisher's site in a new tab

More in Frontier Line and access