Anthropic leaked Claude Mythos 5.5 on its own site, then scrubbed it

On Sunday, November 1, people on X dug into the source of anthropic.com/claude/mythos and found something Anthropic had not shown anyone: the name Claude Mythos 5.5. It sat in a hidden page heading and in the logo's alt text, six times in total, described as "our most capable model class for cybersecurity and biology research."

The visible page did not change. Pricing, access and everything else still described Mythos 5.1, which came out on September 1. An archived copy of the page from October 23 has no mention of 5.5, so the string was added within the last few days.

Then it got a little awkward. At around 17:30 UTC the name quietly disappeared: the header read Mythos 5.1 again and the page source no longer mentioned 5.5. Anthropic has announced nothing and has not commented.

Why it matters. Mythos is Anthropic's locked down line for cyber defense and biology work. Version 5.1 is only available to vetted security and life sciences researchers, while everyone else gets Fable 5.1, the same model with extra safeguards. Earlier Mythos announcements were enough to knock down cybersecurity stocks. The leak also revived talk of Fable 5.5: some users say Anthropic is already testing new checkpoints with select people, and on October 13 a claude-fable-5-5 entry showed up in LiteLLM's Azure price list before being pulled on October 14.

Add the IPO to the mix. On October 26 Anthropic filed for a Nasdaq listing under the ticker ANT, and a big release before the debut would fit nicely. When Mythos 5.5 ships, what it costs and who gets it are all unknown. A header updated early is a strong hint of a launch, not a launch.

Verdict: Anthropic can hide its models from hackers, just not from its own web team.