Anthropic has done what every responsible AI company does when it builds something genuinely dangerous: released it to the public with a nervous laugh and a disclaimer.
Claude Fable 5, a descendant of Claude Mythos (the version that made finance ministers and Pentagon officials visibly uncomfortable), is now available to anyone with an internet connection and a willingness to ignore the fine print. The company’s position is admirably honest, if contradictory: this thing is too advanced for regular users to handle safely, which is exactly why we’re giving it to regular users.
The logic is almost musical. Anthropic’s leadership has spent months explaining to government bodies and tech councils that Fable 5 requires specialized expertise to use responsibly. It can reason across domains most people don’t fully understand. It can generate code, financial models, and arguments with enough coherence to be genuinely useful or genuinely dangerous, depending on who’s holding it. It’s the kind of tool that demands careful handling.
So naturally, they released it to GitHub.
This is not a criticism dressed as satire—it’s just what happened. The tool is real, the capability is real, and the gap between “this is too powerful for unsupervised use” and “here, use it unsupervised” is real too. It’s the modern tech equivalent of releasing a car with a warning label that says the brakes might fail, then shipping it anyway because someone will just reverse-engineer it if you don’t.
The responsible move would have been to keep it restricted. The realistic move was always to release it and hope people read the documentation. Anthropic chose the realistic move, which is fine—just don’t pretend it was the careful one.