Updated 10 November 2026. Anthropic released Claude Mythos 5.2 alongside Claude Fable 5.2 on 9 November 2026; the access rules below come from Anthropic's program pages as of 10 November 2026.
You cannot simply sign up for Claude Mythos 5.2. As of 10 November 2026, Anthropic gives it only to people and organizations verified through two programs: the Cyber Verification Program (CVP) for security work and the Life Sciences Verification Program (LSVP) for biology research. Mythos 5.2 is the same underlying model as Claude Fable 5.2 with more permissive safeguards in those two areas, and it costs the same: $8 per 1M input tokens and $40 per 1M output tokens.
Security teams, and individual researchers for the entry tier, apply in Anthropic's Verification Portal, where the fastest tier has a two-day review target. Life sciences organizations on Team, Enterprise or the API register interest first. If you run Claude Enterprise, Claude Security already scans code with Mythos models without CVP membership.
Claude Mythos 5.2 at a glance
| Question | Answer as of 10 November 2026 |
|---|---|
| What it is | The same model as Claude Fable 5.2, with more permissive cybersecurity and biology safeguards |
| Released | 9 November 2026 |
| Model ID | claude-mythos-5-2 (Amazon Bedrock: anthropic.claude-mythos-5-2) |
| Who can get it | People and organizations approved through the CVP or the LSVP |
| How to apply | CVP: Anthropic's Verification Portal. LSVP: an interest form your admin completes |
| Review time | CVP targets: two days (Defense), two weeks (Red Team), no set target (Specialized) |
| Price per 1M tokens | $8 input, $40 output, $0.20 cache reads, Batch $4 / $20 |
| Context / output | 1M tokens / 128K tokens |
| Platforms | Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
| Data retention | Required, with limited exceptions |
What is Claude Mythos 5.2?
Anthropic's system card describes Fable 5.2 and Mythos 5.2 as two configurations of one new model. Fable 5.2 is for general use and carries extra safeguards that stop some high-risk, dual-use tasks; Mythos 5.2 has more permissive safeguards for cybersecurity and biology. Anthropic calls Mythos 5.2 its most capable model for cybersecurity engineering, and it also powers Claude Security.
Anthropic's reported results for Mythos 5.2, run without Fable's safeguards:
- CVE Rediscovery (finding published vulnerabilities in code from before the fix): 89.6%, against 81.3% for Opus 5.5 without safeguards and 78.3% for Mythos 5.1.
- CyScenarioBench (multi-stage cyber operations): 76.3%, against 67.6% for Opus 5.5 without safeguards.
- De novo binder design (a 24-hour protein design task): 85.1%, against 82.6% for Opus 5.5.
- Terminal-Bench-Science 0.1: 75.3%, against 74.6% for Fable 5.2 with its safeguards on.
In Anthropic's launch post, Mozilla said Mythos 5.2 roughly doubled Mythos 5.1's vulnerability discovery numbers across its scanning campaigns. The system card adds a caution: Mythos 5.2 reasoned about being evaluated more often than earlier models, which makes test results harder to read.
How is Mythos 5.2 different from Fable 5.2?
| Item | Claude Fable 5.2 | Claude Mythos 5.2 |
|---|---|---|
| Who can use it | All Claude API customers and paid Claude plans | Verified program members only |
| Cyber safeguards | Blocks high-risk work such as penetration testing and exploit generation; falls back to Opus 4.8 | Relaxed according to your CVP tier |
| Biology safeguards | Blocks most research biology; falls back to Opus 5 | Relaxed under LSVP grants; CVP leaves them in place |
| Price | $8 / $40 per 1M tokens | $8 / $40 per 1M tokens |
| Fast mode ($16 / $80) | Claude API, Claude Code, Google Cloud and Microsoft Foundry | Verification program members only |
| Rate limit | Its own bucket | Shared with Mythos 5.1 and Mythos 5 |
| Claude Platform on AWS | Listed | Not listed |
One detail matters for mixed teams: each program relaxes only its own area. The CVP changes cyber safeguards and keeps the biology ones, and LSVP grants keep the cyber classifiers. For the general model, see our Claude Fable 5.2 guide.
Who can get access to Claude Mythos 5.2?
Cyber Verification Program
The CVP is for security professionals whose work Claude blocks, such as validating exploits or running authorized penetration tests. Every tier includes Mythos 5.2, Mythos 5.1, Opus 5.5 and Sonnet 5.5, with safeguards set by tier. It runs on paid Claude plans, the Claude Console with billing set up and linked cloud accounts, not on the Free plan.
| Tier | What it allows | Who can apply | Plans | Approved users |
|---|---|---|---|---|
| Defense Access | Security operations, incident response, malware analysis, detection engineering, vulnerability triage and validation | Organizations and individuals, such as open-source maintainers and bug bounty hunters | Pro, Max, Ultra, Team, Enterprise, API | No default limit |
| Red Team Access | Adds authorized penetration testing, red teaming and exploit development | Organizations only | Team, Enterprise, API | Up to 75 |
| Specialized Access | Authorized testing of critical systems such as power grids and interbank transfer systems | Organizations only | Team, Enterprise, API | Up to 25 |
Every tier bans some activity outright, including mass destruction of data or devices, denial-of-service attacks, physical-world damage, extortion and election interference.
Life Sciences Verification Program
The LSVP covers drug discovery, research biology, clinical development and manufacturing work that Fable models block. As of 10 November 2026, its program page says it is in beta for verified organizations on Claude Team, Claude Enterprise and the API, and that Free, Pro, Max and Ultra plans are not eligible. At its September launch, Anthropic said it was not yet on third-party platforms or open to organizations with a HIPAA business associate agreement.
- Standard Use grants cover most life science work, can extend to whole teams and renew once a year.
- High-risk Use grants remove the blocking safeguards for one named project and renew every six months. For Mythos models, Anthropic said these stay limited to a small set of entities with extra vetting while it works with the US government.
How to apply for Claude Mythos 5.2 access
Cyber Verification Program, step by step
- Apply from the organization where you will use it. Approval does not carry over, and Claude and Claude Console accounts are separate organizations.
- Have an Owner or Primary Owner (Claude) or an Admin (Console) submit one application for the organization.
- Prepare verification. Individuals need a government-issued photo ID and a phone with a camera, in case a live selfie is requested; organizations also need their legal name, address and business registration number.
- Describe your security work and how you will use Claude, and confirm your tier's security controls.
- Sign in to the Verification Portal, choose "Cyber Verification Program", apply and accept the terms. Cloud customers first link an AWS account, Google Cloud project or Azure subscription.
What each CVP tier requires
- Defense Access: multi-factor authentication and API keys replaced at least every seven days. From 15 December 2026 it also needs phishing-resistant MFA and no static or long-lived API keys; Anthropic suggests Workload Identity Federation.
- Red Team Access: phishing-resistant MFA, seven-day key rotation, company-domain accounts, organization-controlled devices, outbound network controls and background checks, plus proof such as an ISO/IEC 27001 certificate, a SOC 2 Type II report or a paid bug bounty that has run 12 months. Testing firms can show a CVE credit, FIRST membership or a CREST accreditation.
- Specialized Access: all of the above plus single sign-on and managed devices with malware protection. Anthropic reviews each application with the US government.
Life Sciences program: how to register
Your organization's admin completes the interest form. Anthropic then checks eligibility and, if you qualify, sends details on the formal application, which reviews research credentials, security standards and ethical research oversight. Each grant is tied to the use cases you describe, and Anthropic monitors traffic against them.
What happens after you apply?
- Review: targets are two days for Defense Access, two weeks for Red Team Access (Anthropic says high demand is slowing this) and none for Specialized Access. If you apply high and qualify lower, you get the lower tier while the review continues.
- If refused: you can reapply after seven days with stronger evidence.
- Turning it on: on Pro, Max, Ultra and Team, access switches on automatically. Enterprise admins assign the program to a custom role, and Console admins assign it to a workspace other than the default one.
- Cloud platforms: Mythos access on cloud partners starts about five business days after approval.
- Client work: the CVP covers internal use. Security work for customers needs a separate approval, for example as a managed service.
Anthropic asks approved teams to treat each grant like privileged access and keep each tier in its own workspace or role. Our AI policy template is a starting point for writing down who may use it and for what.
How much does Claude Mythos 5.2 cost?
Mythos 5.2 has the same prices as Fable 5.2 on Anthropic's pricing page, as of 10 November 2026, per 1M tokens:
| Item | Price |
|---|---|
| Input / output | $8 / $40 |
| Cache writes (5-minute / 1-hour) | $10 / $16 |
| Cache reads | $0.20 |
| Batch API | $4 / $20 |
| Fast mode | $16 / $80 |
On Claude plans, Mythos follows Fable's usage rules: Pro accounts use usage credits, while Max and Ultra share up to 50% of their weekly limit between Mythos and Fable. As a hypothetical example, a code review run that reads 20M input tokens and writes 2M output tokens would cost 20 x $8 + 2 x $40 = $240 at list price.
What to use if you cannot get access
- Everyday secure development: Anthropic says you do not need the CVP to review code, patch known issues or triage alerts with generally available models. Applying a known fix, such as the GitLab 19.5.1 critical patch, is that kind of work, and our web application security guide covers the basics.
- Claude Security: Claude Enterprise customers can scan repositories with Mythos models in Claude Security, in public beta, without joining the CVP. Mythos-powered scans run only in the Claude Security app on Claude.ai.
- Fable 5.2: for coding, research and analysis outside offensive security and research biology. Our Fable 5.2 vs Opus 5.5 comparison helps you choose between it and Opus.
- Opus 5.5 through a program: CVP tiers also cover Opus 5.5 and Sonnet 5.5, and Anthropic opened the LSVP to Opus 5.5 for biology research.
- Other labs' gated models: OpenAI's pricing page lists GPT-6 Cyber for its Daybreak cyber program at $20 / $100 per 1M tokens and GPT-6 Rosalind for life sciences at $15 / $75, and Google offers defenders early access through its Fairwind Program. These are vetted too; for open models, see our Fable 5.2 vs GPT-6 Astra comparison.
If your team needs an internal tool built on generally available models, such as a security runbook assistant, tell us what it should do and we will scope it.
FAQ
Can individuals get Claude Mythos 5.2?
Yes, for security work. Independent researchers, open-source maintainers and bug bounty hunters can apply for CVP Defense Access on a Pro, Max, Ultra, Team, Enterprise or API plan, after an identity check with a government-issued photo ID. Red Team and Specialized Access are for organizations only, and the Life Sciences program is currently open only to organizations on Team, Enterprise or the API.
Is Claude Mythos 5.2 free?
No. Joining a program does not change the price: Mythos 5.2 costs $8 per 1M input tokens and $40 per 1M output tokens on the API, as of 10 November 2026. On Claude plans, Pro users pay with usage credits, while Max and Ultra share up to half of their weekly limit between Mythos and Fable. The CVP itself is not available on the Free plan.
How long does Mythos 5.2 approval take?
Anthropic's targets are two days for Defense Access and two weeks for Red Team Access, which it says is running slower because of demand. Specialized Access has no target because each application is reviewed with the US government. On cloud platforms, Mythos access starts about five business days after approval, and if you are refused you can reapply after seven days.
Is Mythos 5.2 smarter than Fable 5.2?
No, it is the same underlying model. The difference is the safeguards: Fable 5.2 blocks many cybersecurity and biology tasks and hands them to Opus models, while Mythos 5.2 allows more of that work for verified users. Where Anthropic reports both on the same test, the scores are close, such as 75.3% for Mythos and 74.6% for Fable on Terminal-Bench-Science 0.1.
Can I use Claude Mythos 5.2 with zero data retention?
Usually not. Mythos 5.2 is a Covered Model, and CVP and LSVP traffic require data retention so Anthropic can monitor for misuse. Organizations that already have a retention exemption for Fable or Mythos models can use the CVP with zero data retention for internal use. Anthropic's Enterprise Frontier Safeguards, now in private beta, will let eligible customers keep monitoring data in their own cloud.
Does Claude Security give me Mythos access?
Partly. Claude Enterprise customers can run Claude Security, in public beta, which scans connected repositories with Anthropic's most capable cyber models, including Mythos, with no CVP membership needed. That access is limited to scans in the Claude Security app on Claude.ai. Fixing the findings in Claude Code uses whatever models your account already has.
Sources
- Anthropic: Introducing Claude Fable 5.2 and Claude Mythos 5.2
- Anthropic: Claude Mythos
- Claude docs: Claude Mythos 5.2
- Claude docs: What's new in Claude Fable 5.2
- Claude docs: Pricing
- Claude docs: Rate limits
- Anthropic: Claude Fable 5.2 and Claude Mythos 5.2 System Card
- Claude Help Center: Cyber Verification Program
- Anthropic: Expanding the Cyber Verification Program
- Anthropic: Introducing the Life Sciences Verification Program
- Claude: Life Sciences Verification Program
- Claude: Claude Security
- Claude Help Center: Why Claude switched models with Fable models
- Claude Help Center: Covered Models
- Anthropic: Introducing Claude Opus 5.5
- OpenAI API docs: Pricing
- Google DeepMind: Fairwind Program
Loading comments...