The upcoming Astra model from @OpenAI may be capable of finding and exploiting zero-day flaws in hardened systems without human help. Axios reports development is slowing while OpenAI adds stricter controls and pauses internal work. (Digg)