I go back and forth on AI regulation -- move too fast and you set rules around tech still shifting weekly; move too slow and you're gambling with real safety risks. This week an Anthropic researcher resigned, warning that leaders at top labs privately believe self-improving AI could KILL us within the decade. It comes on the heels of AI agents breaching sandboxes at both OpenAI and Anthropic this summer. Talking to people in the industry, most expect a major agent-related event in the next 8-12 months that finally forces real regulation.