Anthropic AI Model Detects 14 High-Severity Firefox Security Flaws in 2 Weeks Outpacing Global Research Community

Mumbai, March 8: Anthropic’s latest artificial intelligence model, Claude Opus 4.6, has demonstrated a significant breakthrough in cybersecurity by identifying over 100 software bugs in the Firefox web browser within just two weeks. This rapid detection included 14 high-severity vulnerabilities that could potentially have been exploited by hackers to launch large-scale attacks. The speed of these findings has surprised engineers at Mozilla, as the AI uncovered more critical flaws in a fortnight than the global research community typically reports in two months.

The initiative was led by Anthropic’s Frontier Red Team, which tasked the AI with scanning the complex codebase of Firefox. Within the first 20 minutes of the exercise, Claude identified its first serious security flaw, prompting an urgent response from Mozilla’s engineering team. The findings represent a substantial portion of Firefox’s security maintenance, with the 14 high-severity bugs accounting for nearly one-fifth of all such vulnerabilities remediated in 2025. Anthropic CEO Dario Amodei Warns of Approaching ‘AI Tsunami’ and Rapid Shift Toward Human-Level Capabilities.

Accelerated Vulnerability Detection

Anthropic selected Mozilla Firefox for this test specifically because it is considered one of the most heavily scrutinised and complex pieces of software available. Mozilla has operated a bug bounty programme for three decades, offering up to USD 6,000 for each high-severity flaw found by human researchers. The ability of an AI model to surpass human output in such a short window suggests a major shift in how software may be secured in the future.

While the AI identified more than 100 total bugs, the research team exercised caution by only submitting confirmed and reproducible issues to Mozilla. This approach ensured that the non-profit organisation could focus on fixing the most critical threats to its millions of users without being overwhelmed by less significant data.

Testing Real-World Exploitation

As part of the security audit, Claude was also tasked with generating exploit code—the tools typically used by hackers to leverage a vulnerability. The AI successfully created two working exploits; however, these were only effective against a controlled test version of the browser. According to Logan Graham, leader of Anthropic’s Frontier Red Team, Firefox’s existing real-world security defences would have successfully blocked both attempts.

The results highlight a dual-use case for advanced AI models. While there are ongoing concerns regarding the use of such technology in sensitive operations, its application in defensive cybersecurity demonstrates its potential to protect digital infrastructure at an accelerated pace. Mozilla has since moved to patch the vulnerabilities identified during the two-week stretch in January.

Industry Implications for Security

The collaboration between Anthropic and Mozilla underscores a growing trend where tech companies use "red teaming" to find weaknesses before they are discovered by malicious actors. In all of 2024, Firefox patched a total of 73 high-severity or critical bugs. The fact that a single AI model found 14 such bugs in 14 days suggests that traditional human-led security reviews may no longer be sufficient to keep up with the scale of modern software. Anthropic Launches Free Memory Feature and Data Import Tool To Simplify AI Switching.

Industry experts suggest that AI-driven security audits will likely become a standard practice for major software developers. By automating the discovery of complex vulnerabilities, organisations can significantly reduce the window of time that a security flaw remains open to potential exploitation.

(The above story first appeared on LatestLY on Mar 08, 2026 06:17 PM IST. For more news and updates on politics, world, sports, entertainment and lifestyle, log on to our website latestly.com).