Anthropic Expands Access to Claude for Certified Security Teams, Easing Cyber Restrictions

Anthropic announced on October 6, 2026 that its Cyber Verification Program will now grant approved security professionals broader use of the company's Claude AI models, cutting back several of the safeguards that previously limited how the system could be employed in cyber‑security testing.

The updated program folds the original verification framework into a larger initiative that also incorporates Anthropic’s ongoing Project aimed at strengthening AI‑driven defenses. Participants who pass a vetting process will be able to run more sophisticated prompts, simulate advanced threat scenarios, and explore deeper system interactions without the earlier “sandbox” constraints.

Reducing those restrictions is intended to let red‑team operators probe vulnerabilities with a tool that can generate realistic attack vectors, automate reconnaissance, and suggest remediation steps in real time. Claude’s language capabilities, combined with its growing knowledge of network protocols and exploit techniques, make it a powerful ally for teams tasked with testing corporate defenses.

The move reflects a broader trend among AI developers to balance openness with safety. Earlier this year, several firms opened limited APIs to vetted researchers, arguing that responsible disclosure and hands‑on testing are essential to uncovering hidden risks in increasingly complex software ecosystems.

Security experts have welcomed the change, noting that having a high‑capacity model available under controlled conditions could accelerate the identification of zero‑day flaws and improve incident response playbooks. Anthropic’s decision to tighten the eligibility criteria—requiring background checks, proof of professional affiliation, and adherence to a code of conduct—aims to mitigate misuse while still providing the depth of analysis that seasoned red teams need.

Nonetheless, some observers caution that easing AI constraints may inadvertently lower the barrier for malicious actors if the verification process is breached. Anthropic has said it will monitor usage logs, enforce rate limits, and retain the ability to revoke access should any policy violations be detected.

Looking ahead, the company plans to roll out the expanded CVP in stages, beginning with a pilot cohort of thirty organizations before opening the program to a broader pool later in the year. By offering a more capable version of Claude to trusted security professionals, Anthropic hopes to foster a collaborative environment where AI assists in hardening digital infrastructure rather than becoming a new attack surface.