OpenAI says its newest model can find and exploit unknown software flaws on its own.
TechCrunch reports that OpenAI's newest model, Astra, can find and exploit previously unknown software vulnerabilities on its own, the first model the company says meets its highest cybersecurity risk bar. The capability cuts both ways: powerful for defenders running their own testing, and dangerous in the wrong hands.
Why it matters: AI that can discover zero-days changes the tempo on both sides. Defenders should fold AI-assisted testing into their own security programs now, and assume attackers will too, shortening the window between a flaw existing and being exploited.