Verified access to Claude’s full dual-use reasoning capabilities deepens exploitability analysis, red and purple team engagements, and security control validation for Defensive Networks’ clients.
El Segundo, CA August 10, 2026 –(PR.com)– Defensive Networks, a cybersecurity firm delivering offensive security testing, advisory, and professional services to enterprise clients, has added verified frontier AI Cyber capabilities to a practice built on over a decade of hands-on penetration testing. The firm has been added to Anthropic’s Cyber Verification Program (CVP), giving its testers access to AI reasoning that is restricted by default for cybersecurity work.
The addition fits a pattern. Defensive Networks has consistently moved early on tooling that changes what a security engagement can deliver — across exploitation frameworks, adversary simulation platforms, and AI-assisted testing — on the view that clients are best served by testers working with the same class of capability their adversaries have. Verified frontier AI is the newest addition to that toolkit, not a replacement for what is already in it.
“What has always limited penetration testing is arithmetic — a finite number of tester hours against an environment with a near-infinite number of [theoretical] paths through it,” said Ben Ferguson, Chief Technology Officer of Defensive Networks. “You sample, you prioritize, and you accept you cannot look everywhere. This lets us test at a scale that was never available to a human team alone.”
“Attackers are already running frontier AI against our clients’ environments, and they are not waiting on an approval process,” said Armen Muradian, Senior Security Engineer at Defensive Networks. “Verification means our testers can bring that same depth of reasoning into a client engagement — inside a defined scope, with the client’s authorization, and fully documented.”
The CVP is Anthropic’s application-based review process, scoped to the organization and reviewed by Anthropic directly. Frontier models block high-risk dual-use cybersecurity work by default — vulnerability exploitation analysis, offensive tooling, attack path modeling — because that work is difficult to distinguish from attacker intent. Verification lifts those restrictions for organizations that demonstrate legitimate defensive purpose and exhibit exceptional security standards.
“There is more demand for AI-enabled offensive security than there are firms verified to deliver it,” said Paul Cooney, Chief Executive Officer and President of Defensive Networks. “We wanted to be on the right side of that gap early, and to help close it for the clients who need this work done properly.”
The firm expects to keep expanding the practice as capability arrives. “Coverage at this level has effectively been a luxury good, available to companies large enough to fund a standing bug bounty program,” Ferguson said. “That will not stay true for long. What does not come free is the steering — knowing where to point it, what the output means in a specific environment, and which findings are real. The model scales the search. Our expertise is what makes the result worth acting on.”
Clients interested in AI-assisted offensive security engagements can contact Defensive Networks at www.defensive.com
Contact Information:
Defensive Networks, LLC
Paul Cooney
(310) 955-1600
Contact via Email
www.defensive.com
Read the full story here: https://www.pr.com/press-release/975993
Press Release Distributed by PR.com
Media gallery
