Access to Claude Mythos 5.2

#tools#ai

Claude Mythos 5.2 became available to me on November 15, 2026. This post records what the model is, how it differs from Claude Fable 5.2, and how access to it works, based on what Anthropic had published as of that day.

What it is

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026.1 The two are the same underlying model. They differ in their cybersecurity and biology safeguards, and so in who can use them.

Fable 5.2 Mythos 5.2
Safeguards In place Some removed
Who can use it Everyone Those approved through a verification program
Model ID claude-fable-5-2 claude-mythos-5-2

The rest of the specification is shared.2

Item Value
Context window 1M tokens
Maximum output 128K tokens
Price per million tokens $8 input, $40 output
Knowledge cutoff June 2026

Why it matters for this site

Fable’s safeguards block penetration testing, exploit generation, and vulnerability work on binaries. A blocked request is answered by an older model, Claude Opus 4.8, instead.

That list overlaps with much of what a CTF involves, and with Pwn most of all.

How access works

For security work, access goes through the Cyber Verification Program.3 It has three tiers.

Tier Work it covers Who can apply
Defense Access Security operations, incident response, malware analysis, detection engineering, vulnerability triage and validation Organizations and individuals
Red Team Access Authorized penetration testing, red teaming, adversary emulation, development of offensive tools and exploits Organizations
Specialized Access Authorized testing of critical infrastructure, research into the highest-severity techniques Organizations

Two points are easy to miss.

  • Approval does not remove the safeguards. It adjusts them, by a different amount in each tier. Requests can still be refused, and the Usage Policy applies in full.
  • An individual can apply for Defense Access only. On the Pro, Max, and Ultra plans, it turns on automatically after approval.

Notes

  • Anthropic reports that Mythos 5.2 scored 89.6% on CVE Rediscovery, against 78.3% for Mythos 5.1. This is Anthropic’s own internal evaluation.
  • Anthropic recommends starting most coding work with Claude Opus 5.5, and moving up to Fable or Mythos for the hardest open-ended problems.

Mythos 5.2 wrote the first draft of this post.

Footnotes

  1. Anthropic’s announcement of Claude Fable 5.2 and Claude Mythos 5.2, November 9, 2026. ↩

  2. Claude Mythos 5.2, Claude Platform documentation. ↩

  3. Cyber Verification Program, Claude Help Center. ↩