Anthropic expands access to Claude for security teams with relaxed safety filters

Anthropic is expanding its Cyber Verification Program to provide more security professionals with access to Claude's advanced AI models. This initiative, known as Glasswing, enables vetted organizations and researchers to use Claude with fewer safety restrictions for tasks like vulnerability research, malware analysis, and penetration testing. The program aims to enhance the security of critical software by leveraging AI's capabilities to identify and address potential threats more effectively.

The expansion follows a period of significant activity within the program. Between April and July, partners under Glasswing identified at least 129,000 verified vulnerabilities in software systems. Additionally, Anthropic's own open-source scanning detected 5,500 more vulnerabilities between April and October. These findings highlight the growing role of AI in uncovering security flaws that might otherwise go unnoticed.

More than 33,000 of the identified vulnerabilities have been classified as high-severity or critical, according to Anthropic. The company estimates that the actual number of vulnerabilities may be significantly higher, potentially up to five times the reported figure. This discrepancy suggests that the current data may represent only a fraction of the total issues that could be uncovered with further AI-driven analysis.

The expansion of access to Claude's models could lead to increased costs for organizations seeking to implement robust security protocols. It may also raise concerns about vendor lock-in, as reliance on Anthropic's tools could limit the ability to switch to alternative solutions. Additionally, the governance of AI in security applications remains a key consideration, with ongoing debates about the balance between innovation and oversight.

As the program continues to develop, Anthropic is expected to refine its approach to managing access and safety filters. The company has indicated that future updates may include additional models, such as Claude Opus 5.5 and Sonnet 5.5, further enhancing the capabilities available to security teams. The long-term impact of this initiative on the cybersecurity landscape remains to be seen, but it is clear that AI is playing an increasingly central role in the field.