Anthropic’s Claude AI Finds 22 Security Vulnerabilities In Mozilla Firefox In Two Weeks

Its most recent artificial intelligence model, Claude Opus 4.6, found several security vulnerabilities in the Mozilla Firefox browser as part of a real-world testing collaboration with Mozilla, Anthropic has announced. The San Francisco-based AI firm said the model found 22 vulnerabilities in just two weeks, including a number of high-severity problems. The experiment showed that advanced AI models can actually help cybersecurity researchers spot complex software bugs much faster.

Claude AI Tested on Firefox’s Complex Codebase

Mozilla made a partnership with Anthropic to assess the practical limits of its frontier AI model. The researchers selected Firefox because it is a complex open-sourced project with high security.

ALSO READ: Apple iPhone Fold Design Leaked In CAD Renders Ahead Of Expected Launch With iPhone 18 Pro Series

The model was difficult to find vulnerabilities in on Firefox, the company said.

“We chose Firefox because it's both a complex codebase and one of the most well-tested and secure open-source projects in the world. This makes it a harder test of AI's ability to find novel security vulnerabilities than the open-source software we previously used to test our models,” the researchers said.

How the Experiment Was Conducted

Before looking for new bugs, however, the research team trained Claude Opus 4.6 on a dataset of older Firefox Common Vulnerabilities and Exposures (CVEs). Many of these vulnerabilities from history were successfully reproduced by the model.

After this stage, the AI was then asked to find new vulnerabilities in the newest version of Firefox.

The analysis was initially concentrated on the browser’s JavaScript engine, but researchers subsequently broadened its scope to include other parts of the codebase.

Key Findings From the Experiment

The researchers then verified all reports generated by the AI model before submitting it to Mozilla.

ALSO READ: Google Introduces Android Bench To Rank AI Models For Android App Development

Mozilla Patches Discovered Vulnerabilities

Antropic announced that the vast majority of vulnerabilities discovered by Claude Opus 4.6 this way were patched in Firefox 148 by Mozilla. The last remaining issues should be addressed in future releases.

Ever since the experiment, Mozilla has begun using Claude internally for tasks like security analysis and vulnerability detection.