GitHub said on 9 November that Claude Fable 5.2, Anthropic's newest Fable model, can now be used in Copilot by Pro+, Max, Business and Enterprise customers. GitHub says that in early tests it completed more tasks than Fable 5.1, with fewer steps and tool calls. It is meant for multistep coding in large codebases. Rollout is gradual, billed at list prices.
Anthropic runs safety classifiers, which are automated checks for harmful use. They must see prompts and outputs, so Anthropic keeps that data by default, and does not train on it, GitHub said. The rule covers Fable 5, 5.1 and 5.2; other Claude models stay on zero data retention. Eligible enterprises can skip retention until the calendar year ends. After that, Enterprise Frontier Safeguards apply: automated monitoring, plus options such as storage and encryption keys the customer controls. Admins must switch the policy on; approval alone does not. GitHub gave no eligibility rules and points firms to their account team.
The checks that make the model available also place prompts and outputs with a vendor, so the admin switch is a data decision. Buyers should ask: which work is too sensitive? Who approves the policy? Should we ask the account team about approval now? What is our plan for the year-end move to EFS?