HeadlinesBriefing favicon HeadlinesBriefing.com

Anthropic's AI Finds Bugs Faster Than Microsoft Fixes Them

Ars Technica •
×

Microsoft is engaged in a race against time to patch software vulnerabilities discovered by Anthropic's AI model, Mythos. During a mid-May meeting, Microsoft engineers discussed how Claude Mythos Preview was uncovering bugs at an "unprecedented clip," faster than the company could fix them.

In April alone, Mythos identified 90 critical and 141 important bugs in Microsoft's SharePoint software. By mid-May, even more vulnerabilities were found. Microsoft's engineering manager, Hans Andersen, urged teams to address April bugs quickly, noting a May 31st deadline before other nations or adversaries could potentially leverage similar AI tools.

The situation highlights concerns that AI-driven bug discovery could outpace defensive capabilities, with the potential for low- and moderate-severity bugs to be chained together for devastating attacks. Microsoft maintains its vulnerability triage strategy, prioritizing the most dangerous flaws, but acknowledges the urgency to address customer security.