Anthropic Claude Fable 5

A new frontier AI model that refuses to answer certain questions on purpose — and enterprises are paying double for exactly that.

Anthropic released Claude Fable 5 today. It's their most capable model yet for software engineering, knowledge work, and vision, with its edge growing on longer, more complex tasks. But the real story isn't raw power. It's restraint.

In high-risk domains like cybersecurity, biology, and chemistry, Fable won't answer. It steps back and hands the request to the more conservative Opus 4.8 instead. And it does this without slowing normal work down — Anthropic says 95% of sessions still ran entirely on Fable.

That inverts the pitch we've heard for two years. The old promise was more capability at lower cost. Fable 5 runs $10 per million input tokens and $50 per million output, double Opus 4.8, and part of that premium is governance built into the model itself rather than bolted on afterward.

The market seems to think it's worth it. LG Group signed on the same day to deploy Claude Enterprise across its affiliates.

Here's the strategic read: as AI moves deeper into regulated industries, "safe by default" stops being a compliance checkbox and becomes a feature buyers will pay for. Capability gets you in the door. Trust closes the deal.

Are we entering a phase where built-in governance becomes the real moat in enterprise AI — not benchmark scores?