On an afternoon in mid-May, Microsoft engineers and managers gathered online and at the company's Redmond headquarters for a critical meeting. Project Glasswing was underway, with Mythos—an AI model developed by Anthropic—unearthing software bugs at an unprecedented rate.

The tech giant was racing to patch vulnerabilities before hackers could exploit them. Engineers were in 'a mad dash' to close the gap, as one manager put it. By April alone, Mythos had identified 90 critical and 141 important bugs in Microsoft's SharePoint software. In May, the number of bugs found increased further.

Engineering Manager Hans Andersen urged the group: 'Please, please, please if your org has any April bugs, drive those down.' The deadline was June 31, when adversaries would allegedly catch up with their own tools. One engineer summed it up succinctly: 'So basically you’re saying if it’s released on June 1, then on June 2 the adversaries will have our bugs?'

The meeting highlighted a significant shift in cybersecurity and software development as AI becomes an increasingly powerful tool for identifying weaknesses. The race is on to see who can move faster—Anthropic with their groundbreaking model or Microsoft with their traditional methods.