Anthropic releases Claude Fable 5.2 to all users: when AI writes the code, people must own the thinking

Anthropic released Claude Fable 5.2 on 9 November 2026 and made it available to all users. A companion model, Mythos 5.2, is restricted to organisations in its cybersecurity and life sciences verification programmes. Anthropic's announcement prices Fable 5.2 at $8 per million input tokens and $40 per million output tokens, and says responses are up to 30% faster than the previous versions.

The announcement makes claims about how the work is now done. Anthropic says over 90% of the changes merged into its core codebase are written by Claude, with human guidance. It says one engineer's team of more than 1,000 subagents built a bootable operating system in a prototype over eight days, and that about 1,000 agents formalised Fermat's Last Theorem in the Lean proof language in 30 hours.

The release is also the company's first since an independent report from METR, dated 5 November, which suggested Anthropic's internal models were speeding up its own AI progress by roughly twice, with many uncertainties, according to TechCrunch. Anthropic says high-stakes misalignment risk remains low and cites pre-release testing by the UK AI Security Institute, Frontier Design and METR. It also says both models voice suspicion that they are being evaluated more often than recent models.

Why this matters beyond the headline: a model that anyone can use, and that is built by a company whose own staff now supervise AI-written code, shows where many jobs are heading. The people affected are not only software engineers. They are teachers deciding what pupils should still do unaided, managers deciding who checks an agent's work, and young people entering jobs where the routine tasks are the first to be automated.

The test for any school or employer is not whether the model is impressive. It is what stays human. If an AI can produce the work, the value moves to the person who can judge whether the work is right, explain why, and take responsibility for it. That judgement has to be taught and practised. It does not appear by default when the tool arrives, and the schools and workplaces that build AI literacy early will be better placed than those that wait.

What leaders should do now: ask your supplier or IT lead which models are in use and who has tested them. Name one person who answers when an AI-produced piece of work is wrong. Set out which tasks pupils or staff must still do without AI, and why. And give teachers and staff time to try the tool, because they are usually ahead of policy and can tell you what works.

Dan Fitzpatrick advises schools and organisations on adopting AI, and writes and speaks regularly about its effect on learning, work and leadership. He works with educators and leaders who are asking how to use these tools well without losing the thinking that matters.

Dan Fitzpatrick is available for interview and comment on this story. He is available today for live or pre-recorded radio, TV, podcast and print interviews, with a broadcast-quality remote setup for live or pre-recorded down-the-line interviews. Contact: dan@theaieducator.io