AI tools are improving in cybersecurity. Claude AI identified close to two dozen vulnerabilities in Mozilla’s Firefox browser. In collaboration with Mozilla, Anthropic tested its AI tool for potential weaknesses. A blog post from Anthropic stated that Claude Opus 4.6 discovered 22 vulnerabilities within two weeks, with Mozilla labeling 14 as highly severe. This shows that AI can detect serious security issues swiftly.
While Claude AI excelled in identifying vulnerabilities, it struggled to exploit them. The researchers attempted to have the AI read and write files on a target system, mimicking an attacker’s actions.
Anthropic ran several hundred tests, spending around $4,000 in API credits. Opus 4.6 only managed to exploit vulnerabilities twice. This indicates that discovering vulnerabilities is cheaper than exploiting them. Despite only partial success in developing exploits, the capability is worrying.
The exercise demonstrated that AI tools effectively identify vulnerabilities in open-source projects. Anthropic promotes Claude as a tool to prevent vulnerabilities from being exploited.
Interested in more tech insights? Subscribe to Mashable’s Top Stories and Deals newsletters today.