Anthropic to Pay Up to $15,000 to Hackers Who Find Vulnerabilities in Its AI Systems

Anthropic has announced the launch of an expanded bug bounty program, offering external cybersecurity specialists rewards of up to $15,000 for identifying critical issues in its artificial intelligence systems.

This initiative is aimed at uncovering «universal bypass methods,» which are potential exploits that could consistently circumvent AI security measures in high-risk areas such as chemical, biological, radiological, and nuclear threats, as well as in cyberspace. According to VentureBeat, Anthropic will invite ethical hackers to test its system even before its public release, with the goal of preventing potential exploits that could lead to the misuse of its AI systems.

Interestingly, this approach differs from the strategies of other major AI players. For example, while OpenAI and Google also have bug bounty programs, they are more focused on traditional software vulnerabilities rather than AI-specific exploits. Additionally, Meta✴ recently faced criticism for its relatively opaque stance on AI security research. In contrast, Anthropic’s clear emphasis on transparency sets a new standard in this area.

However, the effectiveness of bug bounty programs in addressing the full spectrum of AI security issues remains debated. Experts suggest that a more comprehensive approach may be needed, involving extensive testing, improved interpretability, and possibly new governance frameworks to ensure that AI systems globally align with human values.

The program will initially launch as an invitation-only initiative (closed testing) in partnership with the well-known platform HackerOne. In the future, Anthropic plans to expand the program, making it open and creating a separate independent model for industry collaboration on AI security.