I honestly think @DarioAmodei is being thoughtful about releasing @AnthropicAI Mythos in a staged fashion. Step one: internal review shows next-level hacking ability ✅ Step two: run a closed beta for a couple of weeks with 50 partners to let them patch any vulnerabilities ✅ Step three: run a thoughtful open beta for logged-in users that is a) audited and b) drops you out of the model if you get too cute Step four: turn off all access when the government asks for a partial shutdown (aka being even more risk-averse) If you replace Anthropic with Microsoft, OpenAI or Google Gemini in the series of circumstances above, you would give them a high five for being thoughtful, right?(@jason)