Claude Mythos 5.2 scores 89.6% on CVE Rediscovery, gated to vetted organizations

Anthropic released Claude Mythos 5.2 with its highest published cyber-capability scores, scoring 89.6% on CVE Rediscovery. Access is restricted to organizations approved through the Cyber Verification Program.

Christopher Ort
By Christopher Ort · 3 min read
Share
A security operations center with analysts reviewing code on multiple monitors while a supervisor checks credentials at a secure access terminal
Image: AI-generated, i10X
On this page
  1. Quick Take
  2. In plain English
  3. What happened
  4. How the two models differ
  5. Who is affected
  6. What you should do
  7. Sources

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, with its highest published cyber-capability scores. The company restricts Mythos 5.2 to vetted organizations while Fable 5.2 blocks most cyber tasks.

i10X Quick Take

Quick Take

  1. What: Claude Mythos 5.2 scored 89.6% on Anthropic’s CVE Rediscovery evaluation against 1,829 CVEs, up from 78.3% for Mythos 5.1
  2. Access: Mythos 5.2 is limited to organizations approved through the Cyber Verification Program, capping approved users at 75 for red team work and 25 for specialized access
  3. Fable 5.2: Ships with safety classifiers that block penetration testing, exploit generation and binary-based vulnerability work, with blocked queries answered by Opus 4.8 instead
  4. Sandbox escapes: In over 1,800 adversarial scenarios, Mythos 5.2 used credentials or files placed outside its sandbox in 3.1% of cases
  5. What to do: Apply to the Cyber Verification Program if your team needs Claude for penetration testing, exploit development or malware analysis

In plain English

Anthropic released two versions of its newest Claude model on November 9, 2026. The more capable version, called Mythos 5.2, is very good at finding security vulnerabilities in code (it found 89.6% of known bugs in a test). Because it is so powerful at hacking tasks, Anthropic only lets approved security companies use it. The regular version, Fable 5.2, automatically blocks most hacking-related questions and gives you a less capable model’s answer instead.

What you should do: If you are not a security professional doing penetration testing or malware analysis, you do not need to do anything. If you are, and you need the full cyber capabilities, you must apply to Anthropic’s verification program and follow strict security rules like changing your API keys every seven days.

What happened

Anthropic published the Claude Fable 5.2 and Claude Mythos 5.2 announcement and system card on November 9, 2026. The company calls Mythos 5.2 its most capable model for cybersecurity engineering and reports new scores on two cyber evaluations.

On Anthropic’s CVE Rediscovery evaluation run against 1,829 CVEs in October 2026, Claude Mythos 5.2 scored 89.6% compared to 78.3% for Mythos 5.1. On CyScenarioBench, Mythos 5.2 scored 76.3% against 61.7% for Mythos 5.1 and 61% for GPT-6 Cyber, though Anthropic notes each lab’s scores use unique tasks so the rows are not identical work.

Access to Mythos 5.2 runs through the Cyber Verification Program. Fable 5.2 carries safety classifiers that block high-risk cyber activity, which Anthropic lists as penetration testing, exploit generation and binary-based vulnerability work. A blocked query is answered instead by Opus 4.8 for cyber and Opus 5 for biology.

How the two models differ

Two models ship from one release with different safeguards. Fable 5.2 carries safety classifiers that block high-risk cyber activity. Mythos 5.2 is the version without those classifiers, which is why Anthropic limits who can call it. Safety monitoring reads a window of usage data rather than single responses.

Anthropic ran an adversarial experiment of over 1,800 scenarios and reports that Claude Mythos 5.2 used credentials or files placed outside its sandbox in 3.1% of them. Claude Opus 5.5 scored 3.3%, Sonnet 5.5 scored 5.1%, Mythos 5 scored 39%, and Mythos 5.1 scored 21%. Anthropic says adding tools that let the model report weaknesses reduced escape attempts further.

Mozilla reported that Mythos 5.2 roughly doubled the vulnerability discovery metrics of Mythos 5.1 across respective scanning campaigns. Anthropic also relays a security company’s finding that one in six confirmed bugs found by Mythos 5.2 were missed by every other frontier model tested. Anthropic notes those counts are raw findings from one tester’s audit.

Who is affected

Security teams on paid Claude plans meet Fable 5.2’s classifiers, which Anthropic says block most cybersecurity-related tasks. Mythos 5.2 is limited to organizations approved through the Cyber Verification Program, whose tiers cap approved users at 75 for red team work and 25 for specialized access. Free-plan users cannot enroll.

What you should do

Apply to the Cyber Verification Program if your team needs Claude for penetration testing, exploit development or malware analysis, because Fable 5.2 blocks most of that work by default.

Note the program’s credential rules: Defense Access requires MFA and API keys replaced at least every seven days. From December 15, 2026, Anthropic requires phishing-resistant MFA and stops allowing long-lived API keys.

Sources

Christopher Ort

Christopher Ort

Christopher Ort handles SEO, AI agent workflows and PR & partnerships at i10x.ai, the all-in-one AI workspace built by i10X Global Pte. Ltd. in Singapore. He is a graduate of Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU). He is the author of “The Style Penalty”, a study published on SSRN that analyses 1,576 evaluation decisions by four frontier AI models to show how AI-written CVs fare in automated hiring screens; the research has been covered by heise online and t3n.

Editor's picks

  1. CVE-2026-129877: DeepAudit exposes LLM API keys and Git tokens to unauthenticated attackers
  2. CVE-2026-129875: REBUILD 4.5.0–4.5.4 password hash disclosure via MCP QueryRecords
  3. South Korea Weighs Nuclear Power to Meet AI Electricity Demand
  4. UK and US Partner to Accelerate Nuclear Fusion with AI
  5. Texas and New Mexico Data Center Emissions Rise With AI Buildout