How Anthropic’s Mythos AI transformed Firefox security in just months

When Anthropic unveiled its Mythos AI model in April, the company warned that the system had discovered thousands of high-severity software vulnerabilities that needed patching before public release. Now, security researchers at Mozilla are detailing how Mythos has reshaped Firefox’s security landscape — uncovering bugs that had remained hidden for over a decade, including critical flaws in the browser’s sandbox protection.

From false positives to real breakthroughs

In a post published Thursday, Mozilla’s Firefox security team reported that Mythos has unearthed a wealth of high-severity bugs, a dramatic improvement over AI security tools from just six months ago. Earlier AI-powered bug finders often overwhelmed security teams with low-quality reports and false positives, making them impractical for real-world use.

But Mozilla’s researchers say the latest generation of agentic AI systems — capable of assessing their own work and filtering out bad results — has turned a corner. “It is difficult to overstate how much this dynamic changed for us over a few short months,” the researchers wrote. “First, the models got a lot more capable. Second, we dramatically improved our techniques for harnessing these models.”

The results are striking: In April 2026, Firefox shipped 423 bug fixes, compared to just 31 in the same month a year earlier. The researchers have also published details on 12 of the bugs, including a 15-year-old error in how the browser parses an HTML element and two unusual sandbox vulnerabilities.

Sandbox vulnerabilities: the hardest bugs to find

The discovery of sandbox vulnerabilities is particularly notable. To find such a bug, the AI must write a compromised patch for the browser, then attack the most secure part of the software with the new code implemented — a delicate, multi-step process requiring both creativity and precision. Mozilla’s bug bounty program offers up to $20,000 for sandbox vulnerabilities, the highest reward available.

Despite the top-dollar bounty, Brian Grinstead, a distinguished engineer at Mozilla, told Bitcoin World that Mythos is finding more sandbox issues than human researchers ever did. “We do get them, but not at the volume that we are able to find with this technique,” he said.

AI finds bugs, but humans still fix them

Despite well-documented progress in AI coding tools, the Firefox team is not yet using AI to fix the bugs it finds. The team does ask AI to generate patches for each bug, but the resulting code usually cannot be deployed directly and instead serves as a model for a human engineer. “For the bugs we’re talking about in this post, every single one is one engineer writing a patch and one engineer reviewing it,” Grinstead said. “We have not found it to be automatable.”

What this means for the cybersecurity landscape

The broader implications of Mythos’s capabilities remain uncertain. One month after the model was previewed, most of the bugs it discovered likely haven’t been patched yet, making it difficult to capture the full scope of its impact. Anthropic has been scrupulous about following responsible disclosure norms, but it’s likely that malicious actors are using similar techniques behind the scenes, even if the models they’re using aren’t as advanced.

Speaking at a recent event, Anthropic CEO Dario Amodei expressed optimism that the new tools would ultimately favor defenders. “If we handle this right, we could be in a better position than we started, because we fixed all these bugs. There are only so many bugs to find,” Amodei said. “So I think there’s a better world on the other side of this.”

Grinstead offers a more measured perspective: “It’s useful for both attackers and defenders, but having the tool available shifts the advantage a little bit to defense. Realistically, nobody knows the answer to this yet.”

Conclusion

Mozilla’s experience with Anthropic’s Mythos model marks a turning point in AI-assisted cybersecurity. The system has proven capable of finding high-severity bugs that eluded human researchers for years, including the most difficult-to-detect sandbox vulnerabilities. While AI has not yet replaced human engineers in the patching process, the technology’s rapid improvement suggests that the balance of power in software security is shifting — with defenders gaining a powerful new tool, even as the same capabilities could eventually be weaponized by attackers.

FAQs

Q1: What is Anthropic’s Mythos model?
Mythos is an AI model developed by Anthropic, designed to detect software vulnerabilities. It was unveiled in April 2026 and has shown significantly improved capabilities over previous AI bug-finding tools, including the ability to assess its own work and filter out false positives.

Q2: How many bugs did Mythos find in Firefox?
Mozilla reported that in April 2026, Firefox shipped 423 bug fixes, compared to just 31 in the same month a year earlier. The researchers published details on 12 specific bugs, including a 15-year-old parsing error and sandbox vulnerabilities.

Q3: Is AI now fixing the bugs it finds?
No. While the Firefox team uses AI to generate patch suggestions, the resulting code usually cannot be deployed directly and requires human engineers to write and review each fix. The team has not found the patching process to be automatable.