Three tiers determine how much Claude can help security professionals

Anthropic has expanded its Cyber Verification Program (CVP) so that approved security professionals can use Claude for more than regular users can. The program now has three tiers: Defense Access, Red Team Access and Specialized Access. How many cyber-related restrictions apply depends on the tier.

Approved users get access to Claude Opus 5.5, Claude Sonnet 5.5, Claude Mythos 5.1 and future models, CSO Online reports.

Three tiers

Defense Access covers defensive work such as incident investigation, malware analysis and vulnerability analysis. Security teams at companies, nonprofits, universities and government agencies can apply. The same goes for critical infrastructure operators, smaller security firms, open source maintainers and individual researchers with a documented history of reporting vulnerabilities. According to Help Net Security, responses are expected to take a few days.

Red Team Access adds authorized penetration testing and simulated attacks. The tier is aimed at internal and public red teams and firms that test the security of others. Ransomware may not be deployed, and systems with high security risk may not be tested. Individual researchers are not eligible for now, and responses are expected to take a few weeks.

Specialized Access has the fewest restrictions and is for verified organizations that test high-risk systems such as aviation systems, power grids and telecom networks. CSO Online also mentions bank networks and government systems. Applications are assessed together with the US government. Existing members of Project Glasswing move to the tier without new approval for the models they already use.

Oversight and data

Organizations must be vetted before they get access. Users may only test systems they are authorized to assess, and access is subject to ongoing oversight. An unnamed expert tells CSO Online that legitimacy depends on authorization, the scope of the target and the conditions of execution, because the steps can resemble those of an attacker.

Anthropic requires data retention in order to monitor for misuse. Enterprise Frontier Safeguards, which is expected to launch in the fall, will, according to Help Net Security, allow approved companies to store data in a cloud infrastructure they control themselves, with continued protection against misuse.

Access is available on Claude Platform, Google Cloud Vertex AI and Microsoft Foundry. On Amazon Bedrock, it is limited to customers with Enterprise Frontier Safeguards.

Between April and July 2026, partners in Project Glasswing found at least 129,000 verified software vulnerabilities, and more than 33,000 of them have been rated critical or severe.