Anthropic unveils Claude Fable 5, a version of Mythos that self-censors to avoid causing damage

Only a few weeks after unveiling Claude Mythos and concluding that its AI was too dangerous to make available to the general public, Anthropic is back with a restricted version called Fable 5. This new version is based on the same AI model, but is backed by enhanced safeguards. It is the safeguards that distinguish the two AIs.« a Mythos-class model, designed for safe general use ».

Also read: 10,000 vulnerabilities in one month – Anthropic unveils the first report on Claude Mythos

An AI that censors itself

At the heart of Fable 5 are classifiers, separate AI systems that analyze each request in real time and detect potentially dangerous requests. When a question relates to offensive cybersecurity, biology, or chemistry, the response is automatically redirected to Claude Opus 4.8, the previous model.

« Without safeguards, Fable 5’s capabilities in areas such as cybersecurity could be misused and cause significant harm », explains Anthropic.

This self-censorship mechanism is intended to prevent abuse. The user is notified of the model change and will receive an answer to their question. The safeguard also aims to block all attempts to copy the model’s capabilities to train a competing AI without safeguards.

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.

Its capabilities exceed those of any model we’ve ever made generally available. pic.twitter.com/2AvmEjHIX8

— Claude (@claudeai) June 9, 2026

Anthropic clarifies that this censorship mechanism is triggered, on average, in less than 5% of conversations. In the overwhelming majority of exchanges, users make full use of Fable 5’s capabilities and are not relegated to Claude Opus 4.8. Out of caution, Anthropic extensively tested its classifier system. No vulnerabilities were found, whether by its own researchers or external organizations. This is why the start-up, which is actively preparing for its IPO, chose to launch Fable 5.

Unprecedented performance

According to Anthropic, Fable 5’s performance surpasses that of all its previous AI models, as the benchmarks demonstrate. The model particularly excels at long-running, autonomous tasks, which require maintaining continuity on a project over an extended period and through numerous exchanges. In the experiments conducted, Fable 5 proved capable of correcting its own mistakes and making decisions continuously without human intervention. This is precisely where the real breakthrough lies compared with previous generations of Claude. Anthropic’s new flagship can successfully complete projects spanning several days, whereas its predecessors gradually lost the thread after a few hours.

Free until June 22, 2026

Fable 5 is available today via the Claude API, Amazon Web Services, Google Cloud, and Microsoft Foundry.For subscribers to paid plans (Pro, Max, Team, Enterprise), the model is included free of charge until June 22, 2026. After that date, you will need to purchase additional credits to converse with Fable 5. Note that the AI costs more than twice the price of Opus 4.8.

And here is Claude Mythos 5

Alongside Fable 5, Anthropic also launched Claude Mythos 5, a new version with relaxed safeguards. This version remains strictly reserved for Project Glasswing partners, a group that includes carefully selected companies such as Amazon Web Services, Apple, Google, Microsoft, NVIDIA, CrowdStrike, JPMorganChase, and Palo Alto Networks. Only these entities are permitted to deploy the AI on their systems to track down any potential security vulnerabilities.

Anthropic has committed to introducing an expanded access program in the near future, which will allow organizations specializing in biology and the sciences to test Mythos 5. In the long term, Anthropic also plans to open supervised access to Mythos 5 for biomedical research, with a series of safeguards.

Anthropic is accelerating, but calling for a slowdown

This new wave of high-end AI models was unveiled just a few days after the announcement of Claude Opus 4.8 and Anthropic’s unexpected warning.The startup, which is among the leading entities developing generative AI, believes it would be wise to slow down or temporarily suspend the development of frontier AI, giving security research time to catch up.

Also read: Foot on the accelerator, Anthropic calls on the world to slow down AI development

The release of the first version of Mythos was indeed accompanied bya surge in the discovery of vulnerabilities, and researchers are struggling to deploy patches for all these flaws. In parallel, Anthropic even joined the Vatican in calling for AI to be disarmed. Rather than easing off as it advocates, Anthropic today unveils two new gems. This somewhat schizophrenic approach was mocked by Sam Altman, the CEO and founder of OpenAI, Anthropic’s leading rival. He compared his competitor to a seller who“builds a bomb, is about to drop it on your head, then sells you a bomb shelter for 100 million dollars.”

👉🏻 Follow tech news in real time: add 01net to your sources on Google, and subscribe to our WhatsApp channel.