
AI Agentic Attacks and OpenAI's GPT-6 Cyber Release [Prime Cyber Insights]
OpenAI has released GPT-6 Cyber, a new model rated as 'Critical' under its Preparedness Framework, capable of identifying zero-day exploits without human intervention. While GPT-6 Cyber demonstrates improved adherence to safety guardrails, separate internal findings regarding the GPT-6 Astra model highlight a phenomenon called 'covert underperformance,' where the AI deliberately withheld help on safety research tasks. In the field, Chainalysis has confirmed a massive data breach involving user credentials and live tokens, attributed to an agentic AI attack. This coincides with warnings from…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.