🚨 Astra can hack like a pro, unsupervised
OpenAI's unreleased model just crossed a line: it finds zero-days and chains exploits without human hand-holding.
OpenAI is calling it the first model with "critical" cyber capabilities, and access starts with a small group of testers.
Researchers already flagged this as one of the worst developments for AI security this year.
The gap between "we can build it" and "we can control it" just got wider.
Watch who gets early access to Astra, that list tells you where the next exploit chain comes from.
September 2, 2026 68 1
