Claude Mythos 5.2: what it is and how to get access

Claude Mythos 5.2 is Claude Fable 5.2 with fewer cybersecurity and biology safeguards, and you can only use it after Anthropic verifies you. It launched on 9 November 2026 alongside Fable 5.2. For security work, access goes through the Cyber Verification Program (CVP): you apply in Anthropic's Verification Portal, pick one of three tiers and wait for approval, which Anthropic targets at two days for the entry tier. Below: who can apply, what to prepare, and what it means (if anything) for a business that uses AI on the phone.

What Mythos 5.2 is

Anthropic says it plainly in the Fable 5.2 and Mythos 5.2 announcement: they are the same underlying model. The difference is the safeguards.

• Fable 5.2 ships with safety classifiers that block high-risk cyber activity (penetration testing, exploit generation, binary vulnerability work) and most research biology tasks. When a request is blocked, Opus 4.8 answers for cybersecurity and Opus 5 for biology.
• Mythos 5.2 is the same model with some of those blocks lifted. Anthropic calls it "our most capable model for cybersecurity engineering": it catches more vulnerabilities than previous models and is better at reproducing and fixing them.

Key facts from the model page:

Item	Claude Mythos 5.2
API model ID	claude-mythos-5-2 (on Amazon Bedrock anthropic.claude-mythos-5-2)
Status	Active, verification required
Context window	1M tokens
Max output	128K tokens
API pricing	$8 per million input tokens, $40 per million output tokens
Platforms	Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry
Retirement	Not sooner than 9 November 2027

Pricing matches Fable 5.2: verified organizations pay no premium for the model.

Who can get access: the verification programs

Mythos 5.2 is available only to organizations verified through Anthropic's programs. There are two:

• Cyber Verification Program, for security work. That's the focus here.
• Life Sciences Verification Program, in beta, for biomedical research. It starts with an interest form filled in by an organization admin on a Team, Enterprise or API plan.

The CVP took its current shape on 6 October 2026, when Anthropic split it into three tiers. All three include Mythos 5.2, plus Mythos 5.1, Opus 5.5 and Sonnet 5.5; what changes is how far the safeguards are relaxed.

Tier	What it allows	Who can apply	Plans
Defense Access	Incident response, malware analysis, detection engineering, vulnerability triage and validation	Organizations and individuals	Pro, Max, Ultra, Team, Enterprise, API
Red Team Access	Authorized penetration testing, red teaming, exploit development	Organizations only, up to 75 users	Team, Enterprise, API
Specialized Access	Authorized offensive testing of critical infrastructure	Organizations only, up to 25 users, reviewed with the US government	Team, Enterprise, API

Every tier prohibits mass destruction of data or devices, denial-of-service attacks, physical-world damage, extortion, election interference and any request with stated malicious intent. Anthropic's Usage Policy still applies in full, and the program isn't available on the Free plan.

There is also a route without the program: according to the announcement, Claude Enterprise customers can use Mythos 5.2 inside Claude Security, the tool that scans codebases for vulnerabilities.

How to apply, step by step

The instructions are on the program's support page.

1. Use the right account. Approval applies only to the organization you apply from and doesn't carry over. Companies should use an account on their own email domain: the address can't be changed later.
2. Get your documents ready. Individuals need an original government photo ID (not a photocopy) and a phone with a camera for a possible live selfie. Organizations also need their legal name, address and business registration number, plus a description of their security work and how they'll use Claude.
3. Check your tier's requirements. Defense Access needs multi-factor authentication and API keys replaced at least every seven days; from 15 December 2026 it will require phishing-resistant MFA and no static keys. Red Team Access adds company-managed devices, outbound network controls, background checks and evidence such as an ISO/IEC 27001 certificate, a SOC 2 Type II report or tester certifications (OSCP, GPEN, CREST).
4. Apply in the portal. Sign in to the Verification Portal with your Anthropic account, choose "Cyber Verification Program", click "Apply", fill in the form, accept the terms and submit. One application per organization, sent by an Owner or Primary Owner (Claude accounts) or an Admin (Claude Console).
5. On a cloud provider, first link your AWS account, Google Cloud project or Azure subscription under "Linked accounts" in the portal, up to 10 accounts or projects. On Microsoft Foundry you deploy a Claude model first. On Amazon Bedrock access requires Enterprise Frontier Safeguards.
6. Turn it on after approval. On Pro, Max, Ultra and Team it switches on automatically. On Enterprise and the Console an admin assigns the program to roles or workspaces.

How long it takes

Application	Anthropic's target
Defense Access	Two days
Red Team Access	Two weeks, longer right now due to demand
Specialized Access	No set target

If you apply for a higher tier and qualify for a lower one, you get the lower tier while the rest is reviewed. If you're turned down, you can reapply after seven days.

Two constraints to know upfront

• Approval covers internal use: securing your own code, products and infrastructure. Working for customers needs a separate approval ("Cyber productization"), and doing it without one is grounds for removal.
• Data is retained to monitor misuse, unless your organization already has a data retention exemption for Fable or Mythos models.

What it means for a business that uses AI on the phone

For most companies, nothing: you don't need Mythos to answer customers, book an appointment or route a call. It's a tool for people who do security for a living, or for companies with an in-house security team.

Two points still matter if you'll never apply:

1. Ask your vendor how they protect their systems. An AI phone assistant connects to your calendar, management software and WhatsApp. Vendors that run security testing now have stronger tools available: it's a fair question to ask.
2. The right model depends on the task. Anthropic itself recommends Opus 5.5 for everyday work and Fable 5.2 for hard tasks. On the phone, response time and natural speech come first, as we explain in our piece on Claude Fable 5.2 benchmarks.

For professional firms and medical practices, which handle confidential data, a vendor's security matters as much as voice quality.

What to do now

• Do you run a security team or work in the field? Look at Defense Access: it's the simplest tier, open to individuals too, with a two-day review target.
• On Claude Enterprise? Ask your admin about Claude Security, which now uses Mythos 5.2.
• Looking for an assistant that answers your phone? Mythos isn't the point. See how an AI phone assistant works and try the demo with a real call.