Anthropic is opening up its most capable security model for code scanning, but only the results, not the model.
Claude Fable 5 is now running inside Claude Security, the company's code-scanning tool, and Anthropic is plugging it into partner products so external security teams can use its output. Users get scan results and findings, not direct access to the underlying model. Anthropic is also putting $35 million in credits toward open-source security work. The company hasn't said which partners get access first or what the credits will actually fund.
This is a familiar move: give away the labor, keep the IP. Vendors have learned that AI-powered security sells better when the AI itself stays proprietary, and keeping Fable 5 behind an API also limits how much a rival or attacker can learn about its scanning techniques.
Whether $35 million meaningfully moves open-source security forward depends entirely on where it lands, and Anthropic hasn't said.