Anthropic Opens Mythos to Hackers. The Price Is Your Logs

Source: Anthropic: Cyber Verification Program · 2026-10-15; Cryptopolitan: Teams must let Anthropic keep data to join cyber program · 2026-10-15

What changed on October 6

Anthropic's announcement describes three tiers, and each one blocks fewer cyber actions than the last. Defense Access covers incident response, malware reverse-engineering and vulnerability analysis and validation. Red Team Access adds authorized penetration testing and red-teaming, but only against systems you're allowed to test. Real-time blocks remain on actions such as deploying ransomware, damaging physical systems or pen testing high-risk safety systems. Specialized Access has the fewest blocks. It is for verified organizations authorized to test safety-critical systems such as power grids, flight systems and telecom networks, and Anthropic reviews each one in depth with the US government. The program now absorbs Project Glasswing, which had given a group of critical-software defenders access to Mythos. Existing Glasswing members move into Specialized Access.

Source: Anthropic: Cyber Verification Program · 2026-10-15

Source: Anthropic: Cyber Verification Program · 2026-10-15

This one sentence is the deal. It applies at every tier. If you want looser filters on Anthropic's most restricted model, Anthropic gets to see what you do with it. For now, the only exception is organizations that already have ZDR access to Fable 5.1 or Mythos 5.1. When Enterprise Frontier Safeguards launches, Anthropic says eligible organizations will be able to keep data in cloud infrastructure they control. The program runs on the Claude Platform, Google Cloud Vertex AI and Microsoft Foundry. On Amazon Bedrock, it's limited to customers eligible for Enterprise Frontier Safeguards.

Source: Anthropic: Cyber Verification Program · 2026-10-15

Who can apply for which tier

1. Defense Access: the broadest tierOpen to security teams at companies, nonprofits, universities and government bodies that defend their own systems. Critical infrastructure operators, smaller security firms, open-source maintainers and individual researchers with a record of reported vulnerabilities can also apply. Anthropic aims to answer within a few days.
2. Red Team Access: organizations onlyOpen to in-house red teams, government red teams and security testing firms. Individual researchers aren't eligible. Anthropic says review takes a few weeks.
3. Specialized Access: invitation-level scrutinyLimited to verified organizations authorized to test safety-critical systems, with in-depth review alongside the US government. Glasswing members are moved here without reapplying for current models.
4. Apply and assignApply through Anthropic's CVP portal. Anthropic verifies applicants and asks for proof of the security controls your tier requires. Admins then assign access to specific workspaces. Existing CVP members keep their settings for older models and are automatically evaluated for the new ones.

Source: Anthropic: Cyber Verification Program · 2026-10-15

Cryptopolitan's report on Anthropic's benchmark data shows how much the tier matters. According to that report, Opus 5.5 blocked every task on the first prompt without the program. Under Defense Access, 46 of 50 trials were blocked at some point. Under Red Team Access, none were blocked.

Source: Cryptopolitan: Teams must let Anthropic keep data to join cyber program · 2026-10-15

Our take: a real opening, with a bill attached

Let's not undersell this. Mythos used to sit behind Glasswing's velvet rope. Now a small security firm or an open-source maintainer can apply for it. That is a real shift, and Anthropic deserves credit for it. But the headline is overrated in one specific way. Defense Access is the tier most people will get, and by Cryptopolitan's account of the benchmark it still blocked nearly every trial at some point. The freedom red teamers actually want comes with organization-only Red Team Access and weeks of vetting. The retention rule is the real fight. Anthropic's argument is simple: you can't police misuse you can't see. That is fair. It is also awkward for a pen-test firm whose prompts contain client network details. Teams that already negotiated ZDR win, and everyone else is asked to trust Anthropic until 'later this fall.' That is a vague promise to plan a security program around.

Source: Anthropic: Cyber Verification Program · 2026-10-15

Optional: see Claude subscription listings from sellers. This is a seller listing, not official access to the Cyber Verification Program and not an endorsement.. That link carries our partner code Sk52g, which is how a purchase is attributed to this site. Promotion code UPD3S.

Claude · Compare price & terms →