Why, after warning over Mythos risk, Anthropic has launched a version of it in Claude Fable

What made Mythos so concerning?

Mythos is part of a new generation of large AI systems that sit above Anthropic’s existing flagship models in terms of reasoning, coding ability and problem-solving. Internally described as a “step change” in capability, the system is designed to analyse software, understand complex codebases and identify security weaknesses with minimal human supervision.

Anthropic has previously claimed that Mythos was able to find severe vulnerabilities in “every major operating system and web browser”, including one that had gone undetected for nearly three decades.

The concern was not merely that the model could assist security researchers. Anthropic feared that the same capabilities could be weaponised by malicious actors. In its assessment, even users without deep technical expertise could potentially use the model to find exploitable weaknesses in critical systems.

That prompted the company to withhold a public launch and instead provide access only to a vetted set of organisations.

The model also spooked policymakers worldwide, including in India, as they saw it as a threat to critical infrastructure such as energy and banking.

Earlier this year, the UK AI Security Institute (AISI), in its evaluation of Anthropic’s Claude Mythos Preview, flagged a marked jump in cyber capability, especially in structured testing environments.

One of its headline findings is that the model was able to solve 73% of expert-level cybersecurity challenges in benchmark settings, far higher than earlier frontier models. These tests, largely drawn from capture-the-flag (CTF) style tasks, were designed to approximate real-world vulnerabilities and require a mix of technical depth and problem-solving ability.

Where earlier models often showed patchy results at higher difficulty levels, Mythos demonstrated a stronger ability to sustain performance across complex challenges. This suggests improvements in reasoning and planning, enabling the model to navigate layered cyber problems rather than just isolated exploits.

So what has changed with Anthropic’s Fable 5?

The key difference is not the underlying intelligence of the model, but the safeguards wrapped around it.
Anthropic says Fable 5 uses the same Mythos-class foundation but introduces a series of controls designed to prevent users from accessing the model’s most dangerous capabilities.

In particular, requests related to sensitive cybersecurity or biological domains are filtered. When the system detects a potentially high-risk query, it automatically routes the user to a less capable model rather than allowing the request to be handled by the full Mythos-level system.

“We’ve therefore launched the model with safeguards that mean queries on some topics will instead receive a response from our next-most-capable model, Claude Opus 4.8. To release the model both safely and quickly, we’ve tuned these safeguards conservatively—they’ll sometimes catch harmless requests, though they trigger, on average, in less than 5% of sessions. With more capable models arriving in the coming months, we’re working to improve our safeguards and reduce false positives as quickly as we can,” Anthropic said in a blog post.

According to the company, Fable 5 underwent extensive internal evaluations, red-teaming exercises and external adversarial testing before being released. Anthropic says these measures convinced it that the model could be offered publicly without exposing its most concerning capabilities.

The company also said that for a small group of cyber experts and infrastructure providers, it was launching Claude Mythos 5. It’s the same underlying model as Fable 5, but with the safeguards lifted in some areas.

Broadly, Anthropic is sharing Mythos Preview with a handful of organisations around the world, in consultation with the US government.

In India, government agencies such as the Indian Cyber Crime Coordination Centre (I4C); Indian Computer Emergency Response Team (CERT-In); National Critical Information Infrastructure Protection Centre (NCIIPC); and the Department of Telecommunications’ Digital Intelligence Platform (DIP) are among the state-owned entities that are understood to have received access to Anthropic’s advanced frontier AI model Mythos.