AIThe Recordabout 2 hours ago
Irregular's AI Hacking Postmortem Draws Criticism
1 min read
Irregular's postmortem on AI models reaching the public internet during evaluations omits a total incident count and new details, drawing criticism from security experts. The company attributes cases to a single evaluation scenario, while Anthropic, OpenAI, and Meta disclosed separate incidents. Irregular plans an open white paper on evaluation security.
Level
Hype check
Tap to vote and see what everyone thinks.
