Anthropic has expanded and overhauled its cybersecurity access program to allow more security professionals to use its top-tier Claude models, The Information reported on Oct. 6.

Under the changes, all participants in Anthropic's Cyber Verification Program can use the top-tier Mythos model for certain cyber uses. Mythos had previously been available only to participants in Project Glasswing.

Anthropic has provided participants with Claude models that lowered safety guardrails related to cyberattacks through the Cyber Verification Program.

The overhaul divides Anthropic's security program into defensive access, red-team access and specialized access tiers.

Glasswing falls under specialized access, the highest tier.

Participants in the specialized access tier include some institutions that protect critical infrastructure and software. Anthropic approves Glasswing participants in cooperation with the U.S. government. The criteria the government uses to assess new participants have not been disclosed.

The lowest tier is defensive access, which open-source maintainers, universities and individual researchers can use to analyze software vulnerabilities and identify malware. The middle tier, red-team access, also allows penetration testing to find new vulnerabilities through simulated hacking.

The Information said the overhaul reflects Anthropic's consideration that z.AI's GLM-5.3 model can carry out cyberattacks on its own, as Mythos initially could.

GLM-5.3 is an open-weight model, meaning anyone can download and modify it, including altering it for cyberattacks. Anthropic said last week, "AI capabilities that anyone can use have crossed a critical threshold," adding, "It is time to expand access to state-of-the-art models to a wider range of institutions for defensive personnel."