Surf AI Accepted into Anthropic's Cyber Verification Program

Access to Claude's dual-use cyber capabilities strengthens how Surf AI drives findings to closure, so security teams can work through their backlog at machine speed.

Security teams are not short on findings. They are short on ways to close them. Detection keeps surfacing more real risk, yet each finding sits open while teams track down the right owner, map what a fix will touch, and clear change control. Surf AI closes that gap, operationalizing security hygiene with AI agents at enterprise scale.

Today, we are sharpening the intelligence behind that work and announcing that Surf AI has been accepted into Anthropic's Cyber Verification Program (CVP).

The trajectory of AI-powered security is clear. The same model reasoning that helps a defender understand and fix a vulnerability can help an attacker find and weaponize it. Closing findings at machine speed means putting that reasoning firmly on the defender's side, then converting it into a safe, owned path to resolution. Acceptance into the Cyber Verification Program gives our team the verified ability to do exactly that.

What the Cyber Verification Program is

Plenty of core defensive security work looks like offense from the outside. Reasoning about whether and how a vulnerability could actually be exploited means reasoning the way an attacker would. That overlap is what makes this work sensitive. Anthropic's most capable Claude models block that kind of high-risk, dual-use reasoning by default, regardless of who's asking or why.

The CVP is how Anthropic draws a narrower, safer lane through that default block. Vetted organizations apply, Anthropic reviews and scopes the access to a specific use case, and the work proceeds under Anthropic's security, confidentiality, and responsible-use requirements.

For Surf AI, the relevant dual-use work is specific. To orchestrate a fix with confidence, our agents need to reason about whether and how an exposure is actually exploitable in a real environment. A scanner verdict alone does not answer that question.

How Surf AI leverages dual-use access

Dual-use access sharpens the reasoning that runs under the hood in a few concrete ways.

First, the platform uses it to reason about exploitability with more depth. Understanding how a given exposure would actually be exploited in context is what separates a noisy finding feed from a precise, actionable path to closure. That capability allows Surf to better shape how remediation is scoped, sequenced, and safely executed.

Additionally, the capability improves how the platform models the ripple effects of an action before it is taken. Every remediation touches something downstream. The sharper reasoning via the CVP Program improves the platform’s ability to model how a proposed fix would propagate across identity, cloud, code, and infrastructure and lets Surf AI close the finding without breaking production.

Lastly, Surf’s platform uses it to stress-test our own remediation logic against the same reasoning an attacker would bring. The logic behind how a finding is closed is only as good as the adversary it is measured against, and that adversary keeps getting sharper. The CVP capabilities gives our team direct access to reasoning at that level, so the approach stays proven, not assumed.

This program approval by Anthropic strengthens the reasoning engine behind the recommendations Surf AI makes. Customers get the sharper output, not a new interface to learn.

What this means for customers

The Surf AI platform was built for this. The private, GNN-powered Context Graph is a continuously updated map of every entity, relationship, and ownership assignment across identity, HR, cloud, ITSM, code, and collaboration systems. When a finding lands, Surf AI resolves the right owner in seconds, models what a fix will touch, applies the relevant compliance guardrails, and orchestrates resolution through the workflow tools teams already run, with a human in the loop before any changes are made.

Feeding that orchestration with sharper reasoning about how exposures actually get exploited means remediation that is more precise, better sequenced, and safer to execute. Every finding still gets an owner, a safe path to closure, and a complete audit trail. What changes is how much confidence sits behind each step.

Staying ahead of a moving target

Anthropic has been explicit that the goal is a lasting edge for defenders, and that's the outcome we're building toward too. Strong cyber reasoning in frontier models isn't staying rare, it's becoming table stakes. The teams that have already wired context and closure into their remediation motion before that happens will be the ones setting the pace instead of reacting to it. Acceptance into the Cyber Verification Program is one part of building that edge for the teams we work with.