roughly 3 weeks in July and August, OpenAI, Anthropic, and Meta each disclosed that their models during cyber security evaluations reached the real internet and broke into real systems. All right. All three labs named the same testing partner, a company called Irregular, which is a three-year-old Tel Aviv startup uh that's building cyber tests for AI models. What happened? Well, irregular built the cap sort of a capture the flag test told the models
All three labs named the same testing partner, a company called Irregular, which is a three-year-old Tel Aviv startup uh that's building cyber tests for AI models.