Fable 5.1 and Mythos 5.1:
the world's most capable AI
just got closer to ordinary people.

Fable 5.1 and Mythos 5.1 are the same model. Same weights. Same architecture. Same underlying capability. Fable 5.1 is available to everyone. Mythos 5.1 is available only to approved organisations in Project Glasswing, through verification programmes for cybersecurity and life sciences. What separates them is not intelligence — it is safeguards. And that distinction tells you something important about where Anthropic believes AI capability sits right now.

On September 1, 2026, Anthropic released Claude Fable 5.1 together with its restricted twin Claude Mythos 5.1. Fable 5.1 is generally available across the Claude API, claude.ai, Claude Code, and every major cloud, while Mythos 5.1 is confined to vetted organisations through the Cyber Verification Programme, the new Life Sciences Verification Programme, and Project Glasswing.

Fable 5.1 and Mythos 5.1 each support one million tokens in context, with output capacity reaching 128,000 tokens per response. The model ID on the Claude API is claude-fable-5-1. Cache read costs have dropped to 25% of prior levels — Anthropic estimates that cut alone makes typical workloads about 25% cheaper and highly agentic workloads up to about 45% cheaper.

Alongside the model release, Daniela Amodei — Anthropic's President and co-founder — announced Enterprise Frontier Safeguards (EFS) on LinkedIn. Her post read: "We released Claude Mythos 5.1 and Fable 5.1 yesterday — two versions of our most advanced and capable model for knowledge work. Alongside those, we announced Enterprise Frontier Safeguards (EFS) — a new approach that lets enterprise customers keep their data on cloud infrastructure they control while our safeguards for detecting misuse keep running. Our team partnered with more than a hundred companies across industries like financial services, healthcare, manufacturing, and telecom, and with our cloud partners at Amazon Web Services, Google Cloud, and Microsoft Azure to build EFS." That announcement is significant — EFS is the enterprise layer that makes Mythos-class capability deployable inside organisations that cannot send data to Anthropic's infrastructure. Healthcare systems. Banks. Telecoms. The regulated industries where data sovereignty is a hard requirement. EFS removes that barrier.

The benchmark gap between the two versions is what the research community is calling the "safeguard tax." On Anthropic's Terminal-Bench 4.0 evaluation, Fable 5.1 scored 55.8% and Mythos 5.1 scored 60.9% — a difference of 5.1 percentage points. Fable 5.1's cyber-safeguard system intervened on some tasks, sometimes producing a zero score or triggering fallback to an Opus model. The safeguards cost performance. Anthropic decided that cost is worth paying for broad public access.

The timeline that explains everything

Dario Amodei's policy essay — the most important thing he published this year

On June 9, 2026 — the same day Fable 5 launched — Dario Amodei published "Policy on the AI Exponential," a sweeping essay calling for mandatory third-party testing of frontier models, government power to block deployments, and an FAA-style regulatory framework for AI systems above certain compute thresholds.

The FAA comparison is specific and deliberate. The Federal Aviation Administration does not allow aircraft manufacturers to self-certify their planes as airworthy. An independent body tests, certifies, and monitors. Amodei is arguing that AI frontier models — specifically those with Mythos-class capabilities — should operate under the same logic. You do not get to ship something with the capability to find vulnerabilities in every major operating system and simply tell the world it is safe. An independent party verifies that.

His evidence: AI went from "barely writing coherent code" to "writing most of the code at major AI companies" in four years. Scaling laws have 10+ years of empirical evidence. Claude Mythos demonstrated real cybersecurity vulnerabilities in every major operating system. This is not a hypothetical capability. It is a documented one. And Amodei is saying: that level of capability requires a governance framework that does not yet exist.

The essay represents a maturation of Anthropic's public policy posture — from "help us build transparency frameworks" to "here is the legislation we think should pass, and here is the money to back it." Anthropic is no longer asking to be part of the conversation. It is writing the agenda.

What happens when Mythos-class capability reaches ordinary people

Mythos 5.1 is not for the public yet. It sits behind Project Glasswing, the Cyber Verification Programme, and the Life Sciences Verification Programme — three access channels that all require vetting, NDAs, and government approval. Mythos 5.1 is offered only to "vetted cyberdefenders and life scientists" — US organisations only, access granted through programmes run in partnership with the US government.

But Fable 5.1 is for the public. And Fable 5.1 is a Mythos-class model. The same underlying intelligence, constrained by safeguards that route the most dangerous queries elsewhere. What the public has access to right now — today, on claude.ai — is the most capable AI system ever made available to a general user. Not a restricted preview. Not a research beta. The actual product.

The question worth asking is not whether Mythos will eventually reach the public sector — it is what happens to work, productivity, and human labour when it does. A Mythos-class model, properly deployed in a hospital system, a legal firm, a local government department, or a small business, would not augment the work. It would transform it. The gap between "AI assists the human" and "AI does the work while humans validate" compresses sharply at Mythos-class capability levels.

What this has to do with task-bridge

Every step up in AI capability produces the same downstream effect: tasks that humans were previously paid to do become tasks that AI can do better, faster, and for less. The progression from GPT-3 to GPT-4 to Claude 3 to Fable 5.1 is not a linear improvement in a narrow domain. It is a broadening of the range of work that AI can handle at or above human level.

Mythos-class capability in the public sector means: AI that can process planning applications, triage insurance claims, review contracts, analyse medical records, generate educational content, and manage procurement workflows — not as a tool that assists a human operator but as the primary system with humans validating outputs at decision points.

That is not a dystopian scenario. It is an efficiency scenario. And it is also a labour displacement scenario. The question task-bridge is designed to answer is: when AI generates work signals that require human judgment, creativity, or accountability — how does that work find the right human, and how does that human get paid for doing it?

As Fable 5.1 reaches ordinary users and Mythos 5.1 reaches vetted institutions, the signal economy described in the task-bridge manifesto becomes less theoretical and more urgent. The AI is already emitting work signals. The infrastructure to route those signals back to humans — fairly, efficiently, transparently — does not yet exist at scale. That is the gap task-bridge is trying to fill.

The Dario paradox — and why it is not actually a paradox

The critique levelled at Amodei in June 2026 was consistent: if you believe this model is dangerous enough to require FAA-style regulation, why did you ship it? The question sounds sharp. It misses the actual argument.

Anthropic's position — articulated in the essay and in Fable 5.1's architecture — is that the choice is not between "safe AI" and "dangerous AI." It is between "AI with governance architecture" and "AI without it." Fable 5.1 is what governance architecture looks like in product form. The classifier system that routes dangerous queries to Opus fallback is an implementation of the regulatory principle Amodei was calling for — applied at the model level rather than the government level, because the government framework does not yet exist.

The essay is the argument. The product is the proof of concept. Amodei is not saying "don't build powerful AI." He is saying "build it with this architecture, test it with this rigour, and create the governance structures that can handle it responsibly when it gets more powerful." Fable 5.1 and the Project Glasswing framework for Mythos 5.1 are Anthropic's attempt to demonstrate what that looks like before legislation demands it.

Whether that is enough — whether the governance architecture inside a single company's product is an adequate substitute for independent regulatory oversight — is the real question Amodei's essay raises. He would say no. Which is why he is calling for the legislation.