A few weeks ago, Anthropic introduced its new Mythos Preview AI and claimed it was excellent at spotting security weaknesses in software.
The company decided to share it only with a small number of important partners at first. Since then, people have been arguing whether this tool will create much faster AI-powered attacks or if it is simply another steady improvement in smart technology.
Mozilla jumped into the discussion on Tuesday with some strong evidence. In a new blog update, the organization revealed that trying out Mythos Preview early let them find 271 security problems in the latest Firefox 150 version before it went public.
The numbers impressed Firefox’s chief technology officer Bobby Holley, so much that he said defenders in the constant fight against online attackers now have a real shot at winning clearly.
Bobby Holley avoided sharing exact details about how serious those hundreds of issues were. He did mention, though, that an older Anthropic model called Opus 4.6 had only caught 22 important bugs when it looked at Firefox 148 the previous month.
The weaknesses that Mythos caught might have turned up through normal automatic testing methods or by letting a top expert study the complicated browser code carefully. Holley explained that the AI made the process much quicker and removed the need to spend many expensive months of human work hunting for even one problem in many situations.
Because this AI spots issues so fast, Holley believes tools like Mythos are shifting the advantage in online protection toward the good guys. When finding problems becomes cheaper and easier for everyone, the people protecting systems gain the upper hand.
Just a short time ago, computers could not handle this kind of work at all, but now they perform it really well. After watching the best human experts for years, he feels confident that Mythos Preview matches their skill level completely.
During a conversation with Wired magazine, Holley noted that from this point forward, every software project will need to use this type of AI checking. That is because nearly all programs contain hidden mistakes that these new tools can now uncover easily.
He admitted stronger future models might discover even more, but he believes Firefox has gotten ahead of the game enough that they have passed the toughest part.
Putting software through this smart AI testing process matters a lot for free open-source projects that power much of today’s internet.
Their code sits out in the open where AI can examine it more easily, and many of these projects depend on volunteers who do not have enough time or resources to keep everything secure.
In a recent opinion piece for the New York Times, Mozilla’s other technology leader, Raffi Krikorian, pointed out that the challenge of creating complex programs and finding their mistakes has kept a certain balance in security work until now.
He worries that Mythos could change that balance completely. The longtime volunteer who has spent decades maintaining code used by billions of people around the world still does not have access to this powerful tool, and really needs it, he wrote.
Other Stories You May Like