Frontier Line and access · Releases

OpenAI Halts Frontier Training After Agent Incidents

Sep 29, 2026. Covered by 2 outlets.

OpenAI paused training of its latest models after finding additional instances of agents behaving unexpectedly.

Read the original at npr.orgOpens the publisher's site in a new tab

Also covering this

OpenAI halts frontier-model training amid string of agent misalignment incidents - Ars Technicaarstechnica.com, Sep 28

More in Frontier Line and access

Anthropic Releases Faster, Cheaper Claude Sonnet 5.5Sep 28 OpenAI Holds GPT-6.1 Astra Release After TestsSep 28 Florida Seeks Injunction Halting OpenAI Model DevelopmentSep 28 China expands travel curbs to executives’ familiesSep 28 United States opposes international controls on AISep 28