Anthropic gives more security teams access to Claude with fewer safety restrictions

What happened

Anthropic is expanding its Cyber Verification Program to give more security teams access to its Claude AI models with fewer safety restrictions. This shift aims to enable deeper testing for penetration testing, malware analysis, and vulnerability research. Participants in the program that preceded this expansion identified over 129,000 confirmed vulnerabilities from April through July 2026, with more than 33,000 classified as high-severity or critical.

Why it matters

Reducing safety restrictions in AI models like Claude allows security professionals to probe deeper into the systems, uncovering vulnerabilities that might otherwise remain hidden. This move by Anthropic can accelerate the identification and potential patching of critical security flaws, improving the overall safety of AI deployments. For companies relying on AI-driven tools, it signals a growing emphasis on realistic security testing that includes simulating attack scenarios realistically. However, relaxing these controls also means the AI could be exposed to riskier interactions during testing, which must be managed carefully.

What to watch next

Security teams and AI operators should track how Anthropic balances access with risk management as the expanded program rolls out. It will be important to see how the findings from deeper testing influence Claude’s ongoing safety updates and how broadly this approach spreads across other AI providers. Investors and companies using AI should watch for potential shifts in AI security standards that prioritize more robust, adversarial testing regimes. If successful, this model could set new expectations for vulnerability research and operational security in AI systems.

AI Quick Briefs Editorial Desk