Anthropic has released a version of Claude Mythos – its advanced cybersecurity AI model – to the public despite previously saying the model was too powerful for public release.

The AI firm announced on Tuesday that it would release Claude Fable 5 to the public with safeguards and limitations set for users, though it did admit “releasing a model this capable comes with risks.

The models’ “capabilities exceed those of any model we’ve ever made generally available,” Anthropic said.

Fable and Mythos are essentially the same Claude model in terms of sophistication, but have different guardrails and access levels. Both models have the ability to work on given commends unattended by a human for longer periods than previous Claude models.

Upon its limited release, Mythos sent shockwaves through the cybersecurity ecoystem as it boasted the ability to discover hidden vulnerabilities and security flaws – an opportunity for firms to shore up their security defences, but equally an opportunity for threat actors to exploit unpatched vulnerabilities.

Anthropic gave preview access to Mythos to about 150 organistions, which will now have access to Claude Mythos 5, which lacks guardrails when it comes to cybersecurity or biology, depending on the organisation’s needs.

Firms using Mythos already reported finding over 10,000 critical security vulnerabilities across their systems.

The small release of Anthorpic’s Mythos had firms and countries scrambling to get a look – despite a coinciding lawsuit with the US Department fo Defense, US government agencies managed to get access to Mythos.

erg.

Recommended reading

• Anthropic Investigates Claimed Unauthorised Access to Mythos
• Anthropic’s Mythos Lands It a White House Meeting
• Is Mythos as Dangerous as Anthropic Claims?

However, many UK banks reported that they were blocked from accessing the Mythos preview despite intents to test their own systems for vulnerabilities.

OpenAI has issued its own cyber model, GPT-5.5-Cyber, and provided some major UK banks access.

It remains unclear how much of a true risk these models bring, or if it is simply a clever marketing strategy to boost interest in the cyber models.

Still, if a model comes with inherent risks, it begs the question on why this would be released to the public which likely does not have the same need for cybersecurity flaw discovery and testing as large organisations.