Microsoft is facing a difficult new security problem: an AI system is finding software flaws at a pace that strains the company’s ability to repair them. The tool, Anthropic’s Claude Mythos Preview, has been used inside Microsoft as part of Project Glasswing, a push to uncover vulnerabilities before hackers or adversarial governments can use similar technology.
The issue is not simply that AI can find bugs. It is that the volume, speed and possible combinations of those bugs may change how software security teams decide what to fix first.
What Project Glasswing revealed
On an afternoon in mid-May, Microsoft engineers and managers met online and at the company’s Redmond, Washington, headquarters to discuss Project Glasswing. According to a recording of the meeting viewed by ProPublica, the group was focused on weaknesses that Mythos was uncovering in Microsoft code.
Anthropic had provided access to select organizations that build software used by regular people, companies and governments around the world. The stated goal was defensive: find and fix vulnerabilities before hackers and adversarial governments like China could use comparable tools for espionage or sabotage.
Inside the meeting, one engineer asked whether Mythos had “live up to the hype that Anthropic claimed it would have had?” A manager answered, “Yes.”
That answer carried immediate operational consequences. The version Microsoft was using was surfacing bugs faster than the company could patch them, and engineers were described as being in “a mad dash” to close the gap.
The SharePoint numbers show the scale
SharePoint, Microsoft’s widely used collaboration software, was one of the clearest examples. A slide from the presentation showed that in April alone, Mythos found 90 “critical” bugs and 141 “important” ones in SharePoint. In the first half of May it found even more.
Engineering manager Hans Andersen pressed teams to reduce the backlog, saying, “Please, please, please if your org has any April bugs, drive those down.” He said they had roughly two weeks “to find as many things and do as much good as we can with this access.”
The urgency centered on May 31. Andersen said that date “is considered the day when the rest of the world will have caught up.” Engineers questioned what that meant in practical terms. One summarized the concern this way: “So basically you’re saying if it’s released on June 1, then on June 2 the adversaries will have our bugs?” The response from others on the call was yes.
That exchange captures the core tension. AI gives defenders a way to discover vulnerabilities earlier, but only if they can patch quickly enough to benefit from the head start.
Why triage is getting harder
Microsoft has focused first on the flaws it considers most dangerous. According to the presentation and Microsoft’s public patch updates, that means prioritizing vulnerabilities classified as critical or important. Internal records indicate the company planned to address “moderate”-severity flaws identified by Mythos later. The documents did not mention “low”-severity bugs.
This kind of triage is common in software security. Teams usually fix the vulnerabilities most likely to cause the greatest harm before turning to lower-ranked issues. Microsoft said its decisions are based on factors including exploitability and customer impact.
But AI-powered bug hunting may complicate that familiar model. Mythos can chain together multiple bugs that build on one another. That means lower-severity and moderate-severity issues, if left open, could potentially become part of a more serious attack path.
Vinh Nguyen, a senior technical adviser to Anthropic and a senior fellow for AI at the Council on Foreign Relations who formerly served as chief AI officer and chief data scientist at the National Security Agency, described the risk plainly: “The problem now is that you can chain four low-level flaws, and that can equal a high severity.” He added, “If you’re Microsoft, the current triage strategy may be underpricing risks.”
Microsoft’s presentation did not mention chaining. A company spokesperson told ProPublica that the technique “has long been considered as part of vulnerability assessment and risk analysis.”
Microsoft says urgency is expected
Asked about the internal presentation and the then-looming May 31 deadline, Microsoft downplayed the idea that the situation was entirely new. A spokesperson said “accelerated targeting and exploitation of new vulnerabilities is not a new phenomenon.” The spokesperson also said the meeting reflected how Microsoft “feels a sense of urgency to help our customers at this time.”
The company added: “What was heard on that call and is true today is that security is Microsoft’s most important priority and teams across the company are prioritizing using AI to discover and remediate vulnerabilities as quickly as possible.”
Microsoft declined to say how many bugs engineers had patched since the presentation. Anthropic declined to comment.
The internal materials reviewed by ProPublica suggested the SharePoint team “will be busy for months,” first dealing with the highest-priority critical bugs and then moving to important ones in August. Microsoft says critical vulnerabilities include so-called worms that can crash systems and spread malware across computer networks. Important vulnerabilities could affect the confidentiality, integrity or availability of user data, as well as the availability of processing resources.
After those categories, the group planned to begin work on roughly 300 “moderate” bugs, according to the presentation.
A broader Microsoft security backlog
The documents did not cover the full range of Microsoft products, but they did point to a broader problem. Since Microsoft started using Mythos earlier this year, the AI system had found hundreds of bugs categorized as critical or important in products including Microsoft 365, Teams and Copilot. As of mid-May, most had not yet been patched.
Andersen described the findings during the meeting: “They’re not profound and exotic, but they’re real.” He added, “And a lot of them are exploitable.”
It remains unclear whether hackers have exploited any specific bug identified by Mythos. The source article says some hackers have used AI to automate attacks and appear to be using Mythos-like technology to find and exploit weaknesses.
There have also been public signs of the patching pressure. Microsoft releases monthly software fixes through “Patch Tuesday.” In June, it released patches for more than 200 bugs, which industry experts called an all-time high. On July 14, the company released patches for more than 60.
The larger lesson is straightforward. AI may help major software makers find vulnerabilities earlier, but discovery is only half of the job. The harder test is whether teams can assess, prioritize and patch flaws before the same type of AI-enabled search becomes useful to attackers.