Anthropic’s New AI Model Can Identify More Software Bugs Than Ever. Microsoft Is Struggling to Fix Them Fast Enough.

In mid-May, Microsoft engineers gathered to address a pressing issue revealed by Anthropic's new AI model, Mythos. The AI had unearthed 90 critical and 141 important bugs in Microsoft's SharePoint software within just April, and even more by mid-May. In a meeting, a manager confirmed that Mythos was delivering bugs faster than Microsoft could patch them, leading engineers into a frantic sprint.

By May 31, Microsoft expected the rest of the world to catch up to the flaws uncovered by Mythos. However, the situation might already be dire. National security experts warned that the window to fix these vulnerabilities before adversaries could exploit them was closing. Microsoft's approach to patching the most dangerous bugs while leaving lower-severity ones for later could be underestimating the risks, as these low-level flaws could be chained together to create more severe vulnerabilities.

Microsoft maintains that its triage strategy considers various factors, including exploitability and impact on customers, and downplays the significance of the May 31 deadline. However, some believe the company's approach may be underpriced in terms of risk, given the increasing frequency of AI-powered bug-finding tools.