Anthropic Launches Claude Fable 5: Mythos-Class AI Goes Public — With Controversy

What Is Claude Fable 5?

Claude Fable 5 represents Anthropic's answer to a challenge that has defined frontier AI in 2026: how do you release a model powerful enough to be genuinely useful for hard problems without also handing a dangerous tool to bad actors?

The model is built on the same underlying architecture as Claude Mythos Preview — which Anthropic previously refused to release publicly after discovering it could identify thousands of cybersecurity vulnerabilities at will. With Fable 5, Anthropic chose a middle path: release the Mythos-class capabilities wrapped in safety classifiers that review and redirect high-risk requests.

The result is a model designed for sustained, autonomous work. Fable 5 can operate on complex projects for days at a time with minimal human oversight — tackling multi-step research, financial analysis, legal document review, and advanced software engineering tasks that previously required constant supervision.

How the Safety Classifiers Work

Fable 5 includes an automated review layer that scans incoming requests touching cybersecurity, biology, chemistry, and health domains. Requests deemed high-risk are silently redirected to Claude Opus 4.8 — a capable model, but not Fable-class. Anthropic designed this as the mechanism that makes it possible to offer Mythos-level capabilities broadly without enabling the worst-case misuse scenarios.

The problem was in the implementation. Within hours of launch, security researchers, scientists, and developers began reporting that perfectly legitimate work was being caught by the classifier. Worse, in many cases they weren't told it was happening — they believed they were talking to Fable 5 but were actually receiving Opus 4.8 responses. Fortune described it as "secret sabotage"; The Register documented multiple cases of harmless prompts being outright refused.

Claude Mythos 5: The Unrestricted Tier

Alongside Fable 5, Anthropic also launched Claude Mythos 5 — the same base model but with the cybersecurity classifiers lifted. This version is available only to vetted partners through Project Glasswing, a collaboration with the US government, and to select biological research institutions. Anthropic describes it as having "the strongest cybersecurity capabilities of any model in the world."

Availability Breakdown

The API model identifier is claude-fable-5. Mythos is claude-mythos-5.

The Bigger Picture

Fable 5's first week illustrates the fundamental difficulty of the safety tradeoff at the frontier. A keyword-and-category classifier bolted in front of a powerful model is a blunt instrument: sophisticated attackers probe its edges, while ordinary users get caught in the blast radius. Anthropic released Fable 5 days after publicly warning that frontier AI was approaching a dangerously capable threshold — and its first week showed just how hard it is to draw a safety line that stops bad actors without punishing legitimate users.

The transparency patch is a start. Whether refinements to the classifier can close the gap between "stops harm" and "doesn't obstruct research" is the question Anthropic will be working to answer in the weeks ahead.