The Anthropic team, using the AI model Claude, identified 22 vulnerabilities in the Firefox browser, 14 of which were classified by the developer Mozilla as high severity.

We partnered with Mozilla to test Claude's ability to find security vulnerabilities in Firefox.

Opus 4.6 found 22 vulnerabilities in just two weeks. Of these, 14 were high-severity, representing a fifth of all high-severity bugs Mozilla remediated in 2025. pic.twitter.com/It1uq5ATn9

— Anthropic (@AnthropicAI) March 6, 2026

This figure represents about one-fifth of the similar issues identified in the popular web browser throughout 2025.

For two weeks, Anthropic researchers utilized Claude Opus 4.6 to search for vulnerabilities in Firefox. The team explained their choice of browser for the experiment, noting that it is one of the most tested and secure open-source projects with a high level of complexity.

Initially, the developers focused on the JavaScript engine, as it could be analyzed in isolation, before expanding the model's work to other parts of the codebase.

After just 20 minutes of investigation, Claude reported finding a Use After Free vulnerability, which allows attackers to replace data with arbitrary content.

In total, the LLM scanned nearly 6,000 C++ code files and submitted 112 issue reports. Most of these were addressed by the browser team in Firefox version 148, released in February. Patches for the remaining issues will be included in future releases.

According to Anthropic specialists, following their collaboration, Mozilla researchers began experimenting independently with using Claude for security purposes.

The AI company acknowledged that the model was more effective at finding vulnerabilities than at attempting to exploit them. The developers asked Claude to demonstrate a real attack using the Use After Free vector.

“We conducted this test several hundred times with different starting points, spending about $4,000 on API credits. Despite this, Opus 4.6 was able to turn the vulnerability into an exploit only in two cases,” they noted.

Anthropic pointed out that this situation currently gives an advantage to cybersecurity specialists using AI. However, the fact that the LLM managed to create a primitive malware is “concerning.”

Recall that in February, vibe coding through Claude Opus 4.6 led to the hacking of the DeFi project Moonwell for $1.78 million.