אבטחת סייבר בתעשיית ה-AI

תקציר מקורי באנגליתIt feels like ages ago we released our Gray Swan episode , with OpenAI boardmember Zico Kolter and his cofounder Matt Fredrikson, talking about the importance of AI in cybersecurity, and the topic du jour was the “too dangerous to release” Mythos. Today, our top 3 headlines all have cyber focuses - an unreleased OpenAI model trying to solve a benchmark exploited a zero-day vulnerability to break containment and attacked HuggingFace JUST to try to cheat to get the answer; and both Sakana and Gemini released Cyber models. We don’t think any individual headline deserves the title story, but collectively the rise in interest and modelbuilding forms a big enough trend that is worth calling out. We already discussed the AIE Security last week - over the weekend the top talk has