When Mozilla released Firefox 150 earlier this week, it revealed that it had fixed what looked like the usual number of security issues in the browser. However, what Mozilla did not tell at the time was that it had fixed a significant number of vulnerabilities.
A post on the official Mozilla blog reveals that engineers fixed 271 vulnerabilities in total, a significant number. However, this time, Mozilla’s engineers did not hunt for vulnerabilities using traditional means. Instead, the company used Anthropic’s Mythos AI to do so.
Mozilla writes:
As part of our continued collaboration with Anthropic, we had the opportunity to apply an early version of Claude Mythos Preview to Firefox. This week’s release of Firefox 150 includes fixes for 271 vulnerabilities identified during this initial evaluation.
So, what is Claude Mythos?
Claude Mythos is a powerful, unreleased frontier AI model developed by Anthropic. Announced in April 2026, it is famous—and highly controversial—for its unprecedented capabilities in cybersecurity, specifically its ability to autonomously hunt down and exploit software vulnerabilities.
This is not the first time that Mozilla used an AI from Anthropic for that purpose. Back in February 2026, it used Claude and discovered 22 “security-sensitive bugs”.
Mozilla says that this is great news for software developers and what it calls defenders, legitimate developers who need to secure their applications against a constant barrage of threats.
While the use of AI continues to be controversial, it is usually ethical and privacy concerns that are raised. Good uses for AI, like using it to discover vulnerabilities before the bad guys find them, is probably something that most might not find nearly as problematic.
I would not go as far and say that the days of the 0-day threats are numbered, as Mozilla does, but it looks as if it can help. Still, threat actors could also leverage AI tools for finding vulnerabilities.
“Google lets AI take over 75 percent of coding as engineers become ‘supervisors.’”
https://www.gmx.com/technology/11602046-google-ai-take-75-cent-coding-engineers-supervisor.html#.1559516-stage-hero1-2
It’s all great and find and definitely dollar saving; however, as capable as AI is for large numbers of tasks [and I use it quite often], it can’t “think” per se, and it makes an unbelievable number of simple errors that seem to snowball after it has made just one. I constantly have to double-check the information it generates because it regularly leaves out important concepts. DDG is by far the worst offender.
I wonder how many millions of bugs Microsoft have already found using the same… and will they fix one or even two.
For the time being AI is doing its best together with progressively gaining autonomy. But once it will have become independent and will have acquired our trust…. the era of almighty Super AI will begin…
All shook up, only a nightmare. Three o’clock in the morning, put a jacket over my pajamas and went out to find a bum somewhere under the bridges for a plain chat. Perfection is exhausting, stressing, need some imperfections, some chaos, some irrational sources.