Anthropic launches Claude Fable 5, a safer AI model

Anthropic has launched Claude Fable 5. This new artificial intelligence model matches the level of Claude Mythos, but it will refuse to carry out cyberattacks and will not answer questions relating to biology or chemistry.

Anthropic will not give you access to Claude Mythos, its powerful AI that is particularly effective at finding security weaknesses. Instead, it has just released another model, Claude Fable 5. It has the same capabilities as Claude Mythos 5, but is designed to decline requests when an answer could be dangerous. Anthropic says Claude Fable 5 outperforms every other AI model in all the performance benchmarks it uses, whether for software development, administrative work, image analysis or scientific research.

Claude Fable 5: a powerful AI for the public

It could therefore be the most powerful AI currently available to the general public. Initial tests have been highly encouraging. For instance, Stripe engineers say that, with Fable 5, they reduced months of engineering work to just a few days. The new AI also performs especially well in finance. Anthropic also states that Fable 5 can remain focused during longer tasks, “and improves its results through its own notes”.

Claude Fable 5 is available to the public from today. This AI, whose performance is comparable with Mythos, will be included with Claude Pro, Max and Team subscriptions until 22 June. After that date, credits will need to be purchased to access it. However, once Anthropic has enough computing capacity, it plans to make Fable 5 permanently available to Claude subscribers.

An AI capable of conducting cyberattacks or creating a dangerous virus

To make Fable 5 available to the public, Anthropic had to introduce significant safeguards against malicious uses. “Mythos-type models excel at discovering and exploiting software vulnerabilities. They can therefore make cyberattacks considerably easier and cheaper to carry out. These models also demonstrate strong automated hacking capabilities”, Anthropic acknowledges. On Fable 5, however, a classifier system prevents the AI from advancing such tasks.

Restrictions on biology and chemistry requests

Anthropic also had to take extreme steps to ensure that Fable 5’s advanced biology and chemistry capabilities could not be used maliciously, for example to develop a virus. When a user asks questions in these areas, they are automatically redirected, in most cases, to Claude Opus 4.8, a less powerful model.

Anthropic will nevertheless continue providing the unrestricted version of this AI, Claude Mythos 5, to organisations that use it to defend themselves against cyberattacks. The list of organisations able to benefit from Mythos will also continue to grow. In addition, Anthropic has announced that it will offer Claude Mythos 5 to organisations working in biomedical research as well.