What's this about?
A recorded OpenAI presentation from the Black Hat security conference walks through a real AI-driven security incident nicknamed "The Hugging Face Incident." The security community — Simon Willison and Ethan Mollick among them — is treating it as a rare, concrete case study of how an agentic attack actually unfolds step by step, rather than the usual hand-wavy "AI could be dangerous" talk.
What is Black Hat?
Black Hat is one of the world's biggest professional cybersecurity conferences, where researchers publicly present real vulnerabilities, attack techniques, and incident post-mortems. A talk landing there means it's been vetted as technically serious. When a frontier lab like OpenAI presents an AI-related breach there, it moves the conversation from speculation to "here's exactly what happened and how."
What's the community saying?
- "Watch it even if you don't do security" — Mollick's pitch is that this is the clearest public demonstration yet that AI has gotten good enough to matter for real attacks.
- Building the attack timeline — Willison and others are reconstructing the incident frame-by-frame to understand the exact chain of steps.
- Ties into the week's theme — read alongside the Astra "critical cyber" announcement and Anthropic's injection defenses, people see a single story: agentic security is now the frontier.