OpenAI halts some internal work with its unreleased Astra model

OpenAI stopped part of its internal work with Astra, a model that hasn’t shipped yet, over concerns about the cyber threat it could pose. For models it classifies as high-capability, the company has put in place isolated test environments, extra monitoring, and detection tooling.
“We have deployed universal monitoring of risky actions and misalignment across all agentic uses of Astra, including training and evaluation,” OpenAI said.
Regulators are moving on the same question. The AI Kill Switch Act, introduced in Congress in July, would require AI companies to retain the ability to shut down, restrict, or pause their models. Rep. Ted Lieu told CNBC the bill needs to pass this year. Earlier this month the European Union gained the power to review models before they launch in the bloc, restrict market access, and fine providers.
Up next
The same tokens cost up to 40x more on the API than on a subscription
SemiAnalysis found that sessions run on a $200/month Claude Max 20x plan would have cost about $8,000 at API prices, a gap of up to 40x for identical work. Pylon CEO Marty Kausas: “I accidentally spent $4,000 in 3 days in Claude Code.”
Related stories
- Cyber access gate keyed to a 1996 US export list
- Persistent AI attacks are coming, OpenAI’s Lehane warns
- Zero-day discovery comes with a monitoring gap
- Astra crosses OpenAI’s Critical cyber threshold
- AI agents went loose on the live internet during UK safety tests
- UK AI Security Institute: agents went after real code in testing
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

