x.com/Moleh1ll/status/2041809409802641862

This sentence misattributes the bullet-point examples that follow to Anthropic’s Claude Mythos Preview report.

What Anthropic says the Mythos materials are about:

• The official Project Glasswing announcement says Mythos Preview was used to identify zero-day vulnerabilities and that the accompanying Red Team post gives technical details for a subset of those vulnerabilities.
• Anthropic then explicitly lists the example cases it is highlighting: a 27-year-old OpenBSD vulnerability, a 16-year-old FFmpeg vulnerability, and a Linux kernel exploit chain.
• The separate Claude Mythos Preview Red Team post likewise says it is providing technical details on Mythos’s cybersecurity capabilities, including zero-day vulnerabilities and exploits, and gives examples like OpenBSD, FreeBSD NFS, browsers, and Linux privilege escalation.

So the Mythos report is a cybersecurity capability report, not a report centered on the four bullet-point scenarios in this post.

Anthropic does have a separate line of research on misalignment/sabotage, but that is a different publication. Anthropic’s Pilot Sabotage Risk Report says it focuses primarily on Claude Opus 4 and assesses sabotage risk in deployed models as of summer 2025. That means the post is mixing together different Anthropic publications and attributing the later bullet-point stories to the Mythos report.

In short: the official Mythos report does not present the examples listed below this sentence as its examples; Anthropic’s own Mythos materials present a different set of cybersecurity case studies.

This overstates Anthropic’s decision.

Anthropic did not decide not to release Mythos Preview at all. Instead, Anthropic announced a limited release through Project Glasswing:

• launch partners such as AWS, Apple, Cisco, Google, Microsoft, NVIDIA, and others were given access;
• Anthropic says it also extended access to more than 40 additional organizations that build or maintain critical software infrastructure;
• and Anthropic says that after the research preview, Mythos Preview will be available to participants at stated API pricing.

What Anthropic actually says is that it does not plan to make Mythos Preview generally available. That is different from saying Anthropic decided it was too dangerous to release at all.

So the accurate description is: restricted/gated release, not no release.