Anthropic's Mythos AI finds Microsoft vulnerabilities faster than they can be fixed

Microsoft received early access to Anthropic's Mythos AI through Project Glasswing. The AI discovers vulnerabilities in Microsoft products faster than engineers can fix them. In April alone, Mythos found 90 critical and 141 serious flaws in SharePoint. Microsoft is prioritizing only critical and important vulnerabilities, leaving medium-severity and low-severity bugs for later or unaddressed. However, Mythos can chain low-severity vulnerabilities into serious attacks, a technique Microsoft acknowledges. Microsoft has not yet fixed many critical and important vulnerabilities in products like Microsoft 365, Teams, and Copilot. The company's security team is understaffed, and the problem is expected to worsen as other AI models emerge.