Anthropic is expanding its Cyber Verification Program, adding new access tiers to the system it uses to vet who gets to work with its models on security-related tasks. The company says the tiered structure is meant to widen the pool of defenders who can use its technology while keeping a lid on the risk that the same capabilities get turned against the people they're supposed to protect.
The news was reported by Crypto Briefing. No specific release date was given for when the new tiers take effect.
What the Cyber Verification Program actually does
Anthropic's Cyber Verification Program is the gatekeeping layer between its AI models and the cybersecurity work those models can be pointed at. The idea is straightforward enough: the same model that helps a security team hunt for vulnerabilities can help someone else find those vulnerabilities first. Verification is how Anthropic decides who's on which side of that line.
Adding access tiers means the program won't be a single yes-or-no checkpoint anymore. Different levels of access presumably come with different levels of scrutiny — though Anthropic hasn't laid out the specific criteria for each tier in what's been published so far.
The dual-use problem Anthropic is trying to crack
Dual-use technology is the industry term for tools that cut both ways, and few things cut both ways like AI applied to security. A model that can scan a codebase for exploitable bugs is also a model that can tell an attacker where to push. There's no clean way to separate those two functions once the capability exists.
Anthropic's answer, at least in part, is tiering. Let more legitimate researchers and defenders in, but route them through a verification process that scales with what they're asking to do. It's an attempt to avoid the two obvious failure modes: locking out everyone and leaving defenders without good tools, or letting anyone through and hoping for the best.
That's a real tension, and it doesn't have a tidy solution. The company is betting that a layered system beats a flat one.
Why the timing matters
AI companies have spent the better part of this year arguing about how much access to give security researchers without handing over something dangerous. Anthropic's expansion is a concrete step in one direction — more access, more structure — at a moment when the rest of the field is still debating the principle.
Whether tiered verification becomes the norm or stays an Anthropic-specific approach is an open question. The company hasn't said how many applicants it expects to move through the new tiers, or what happens when someone fails verification but has a legitimate use case.
What's still unspoken
What remains unclear is the operational detail. How many tiers are there? What separates one from the next? Does a lower tier mean less capable model access, or the same access with heavier monitoring? Anthropic hasn't disclosed any of that publicly.
Those omissions aren't unusual for a program still being built out, but they matter for anyone trying to figure out whether the expansion is a meaningful opening or a modest re-labeling of the existing process. The company framed the move as a way to strengthen cybersecurity defenses, and that's the stated goal. The mechanics behind it are still under wraps.
The next thing to watch is whether Anthropic publishes the tier criteria — and whether other AI labs follow with verification programs of their own.
