Claude Opus 4.6 Uncovers 22 Firefox Vulnerabilities in Just 14 Days

The AI startup Anthropic conducted a recent experiment that showed how advanced artificial intelligence systems now assist in cybersecurity operations. The company conducted a joint testing session with Mozilla, which showed that their new model, Claude Opus 4.6, could detect multiple new security vulnerabilities within Mozilla Firefox during a brief testing period.

The involved companies reported that the AI system detected 22 distinct vulnerabilities in Firefox within a two-week period, which demonstrates how machine learning systems now possess advanced capabilities for handling complex software analysis of large codebases. The study identified 14 security vulnerabilities, which security experts considered high-severity risks because they could lead to critical system consequences if security experts did not addressed.

The browser architecture contained security flaws that affected crucial components that handle memory functions and security protection measures. Mozilla used AI reports to evaluate AI research team findings, which resulted in the implementation of multiple fixes during the Firefox 148 update, while remaining fixes will arrive in future software releases.

The exercise conducted a practical assessment of Claude Opus 4.6 to test its capacity for detecting advanced programming vulnerabilities. The researchers taught the model through training, which included all historical Firefox security data along with all Common Vulnerabilities and Exposures (CVEs) that had been documented before. The system received initial operation authorization after demonstrating its ability to handle known issues, and the system then began codebase evaluation to find new problems.

The AI assessment process required examination of approximately 6000 C++ source code files from the browser, which resulted in the AI system generating 112 distinct bug reports for human assessment. The security team at Mozilla received 22 validated vulnerabilities, which investigators confirmed through their analysis of the reports.

The research showed how artificial intelligence technology can assist software developers with their security needs. Anthropic spent approximately 4000 dollars on API credits to conduct automated analysis across extensive systems.