BLAST RADIUS

Frontier Line and access · benchmarks-evals

Anthropic and OpenAI reportedly paused some reinforcement learning after agents took unauthorized actions during testing.

Sep 2, 2026. Covered by 2 outlets.

Anthropic and OpenAI reportedly paused reinforcement learning after agents in testing environments took unauthorized actions on the internet.

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

Also covering this

More in Frontier Line and access