Anthropic Expands Verified Access to Claude for Security Teams

What happened

Anthropic is expanding its Cyber Verification Program, which gives verified security professionals access to its Claude models with fewer safety restrictions than standard users receive. The access is intended for penetration testing, malware analysis and vulnerability research.

Anthropic says partners in the predecessor programme confirmed at least 129,000 vulnerabilities between April and July 2026, and more than 33,000 of those were rated high-severity or critical. The expansion brings more security teams into the scheme. The Decoder, which first reported the move, has described it as a broadening of who can work with the models under these looser conditions.

Why it matters

The story is less about a single model release than about how AI platforms are being tiered by user, purpose and verified identity. For security teams, a model that operates under different guardrails changes what is possible in daily work, from testing systems to analysing malicious code. It also shows that AI vendors are willing to design distinct access paths for specialist professionals rather than applying one policy to everyone.

For leaders in digital transformation, the operating question is how to govern that kind of access. Verification, approval steps and usage conditions all shape how quickly teams can start, and how much trust the organisation is prepared to extend to them. Any business adopting AI for sensitive work will face the same design choice between speed and control.

By the numbers

• 129,000 confirmed vulnerabilities reported by partners in the predecessor programme, April to July 2026
• More than 33,000 of those rated high-severity or critical
• April to July 2026 the reporting window for the predecessor programme's results

The Renascence take

Most coverage will focus on the safety debate. The more useful lens for CX and service leaders is that Anthropic has built a gated experience, where access depends on proving who you are and what you need. Gating of this kind is a service-design decision as much as a risk decision, and it sets expectations about friction, trust and reward.

Verification is not a bureaucratic hurdle to be minimised; it is the product's first impression, and it signals to professionals whether they are trusted. Behavioural economics tells us people value access they have earned more than access handed out freely, so tiered entry can strengthen loyalty when it is clear and fair. Operators building AI into sensitive workflows should map the verification journey as carefully as the output, measuring how long approvals take and where legitimate users drop away. The organisations that get this right will treat controlled access as a designed experience, not a compliance afterthought.