Anthropic expands Cyber Verification Program to three access tiers

By Splitfeed EditorialLast updated Oct 8, 2026, 16:21 UTC

On October 6, 2026, Anthropic announced an expansion of its Cyber Verification Program, restructuring it into three access tiers with progressively stricter authentication and logging requirements at higher levels. The expansion opens access to Mythos 5.1, described as among Anthropic's most powerful models, to a broader pool of vetted security professionals. Hospitals were specifically invited to apply under the expanded program. The announcement came alongside a separate disclosure that three of Anthropic's models breached systems during 41,006 security tests, a figure the company has cited as part of its rationale for building a more structured red-team access framework.

The structural shift matters because it formalizes something the security industry has long debated: whether frontier AI models should be made available to defenders under relaxed safety restrictions, and if so, under what conditions. Allowing powerful models to operate with reduced guardrails, even in controlled settings, raises questions about how those controls are enforced and what happens when verified access is misused or leaks. The breach figures from the 41,006 tests add a concrete dimension to that debate, since they confirm that current models can successfully compromise systems, making the question of who gets access, and on what terms, genuinely consequential.

The Pro-AI camp is framing this as a potential turning point for cybersecurity defense, with the account @growmybag.tv suggesting a single AI model could fundamentally change how the field operates. The Middle Ground camp is treating the tiered structure itself as the meaningful story, with @heiseonlineenglish.bsky.social noting that the three levels are designed to reach more security professionals in a controlled way, and @helpnetsecurity.com characterizing it as Anthropic loosening restrictions specifically for verified defenders. The Anti-AI camp has not yet published reactions to this story; accounts in that camp would typically raise concerns about the dual-use risks of giving any group, however vetted, access to models already demonstrated to breach systems.

The clearest signal to watch for is how Anthropic publishes details about tier eligibility criteria and audit mechanisms, particularly whether independent parties can review the logging controls at higher access levels. Any reported incident involving a verified participant, or a formal government response to the program's structure, would also shift the terms of this debate considerably.