Claude AI Exposes 22 Firefox Vulnerabilities in Just Two Weeks
Recent advancements in artificial intelligence have significantly enhanced the capabilities in the field of vulnerability research. AI systems now extend beyond basic code assistance to fully autonomous threat detection.
Recent advancements in artificial intelligence have significantly enhanced the capabilities in the field of vulnerability research. AI systems now extend beyond basic code assistance to fully autonomous threat detection.
The discovery rate achieved by AI surpasses traditional manual research methods, with AI identifying more vulnerabilities in a single month than human researchers did in any month of 2025. Among these, fourteen were classified as high-severity, accounting for nearly 20% of the critical Firefox flaws patched in the previous year.
A collaboration between Anthropic and Mozilla employed the Claude Opus 4.6 model to independently discover 22 security flaws in Firefox within a two-week period in February 2026. The model was initially tested on the CyberGym benchmark before being applied to the live Firefox codebase.
The research focused on the JavaScript engine of the browser due to its extensive attack surface, which often processes untrusted external code. Within twenty minutes, Claude identified a critical Use After Free memory vulnerability, a type of memory corruption that allows data to be overwritten with malicious payloads. By the project's conclusion, the AI had analyzed nearly 6,000 C++ files and submitted 112 unique reports to Bugzilla.
While the AI is proficient in identifying zero-day vulnerabilities, converting these into functional exploits remains inefficient. The model was tasked with creating primitive exploits to read and write local files on a target machine.
Despite hundreds of automated attempts costing $4,000 in API credits, successful breaches occurred only twice. This indicates that identifying vulnerabilities is currently more cost-effective than exploiting them. Moreover, these attacks were conducted in a controlled testing environment that bypassed Firefox's standard sandbox protections. In real-world scenarios, the browser's defense mechanisms would likely mitigate these AI-generated exploits.
Metric Technical Discovery Details
Recent advancements in artificial intelligence have significantly enhanced the capabilities in the field of vulnerability research.
Target Component Firefox JavaScript Engine (C++ codebase)
Initial Discovery Use After Free memory corruption
Scan Volume 6,000 C++ files analyzed
High-Severity Flaws 14 vulnerabilities
Exploit Success Rate 2 successful local file read/write attacks
Mozilla promptly triaged the submissions and released security patches to millions of users in the Firefox 148.0 update. As language models evolve into sophisticated vulnerability researchers, it is essential for defense teams to modernize patching infrastructures accordingly.
Security teams employing AI for bug hunting should implement task verifiers to authenticate findings. These verification mechanisms enable AI to continuously test its own work, ensuring that a proposed patch effectively eliminates the vulnerability while maintaining core functionality.
Researchers should provide maintainers with actionable data to streamline the disclosure process. The Firefox security team emphasizes the need for the following elements in AI-generated bug reports:
Minimal test cases to isolate the specific crashing input. Detailed proofs-of-concept demonstrating the exploit execution path. Verified candidate patches to expedite the final remediation timeline.
Based on reporting by GBHackers.