Microsoft launches AI cybersecurity model and platform

Microsoft announced a new suite of AI-powered cybersecurity tools on Monday, July 27, 2026, headlined by its first in-house model built specifically for finding software vulnerabilities and a new platform designed to automate threat response. The announcements, made at an event in San Francisco, position the company to compete more directly with AI security offerings from Anthropic, Google, and OpenAI.

The core of the new offerings is MAICyber1Flash, a compact AI model developed to identify challenging vulnerabilities in complex codebases. Microsoft stated the model underwent review by its internal AI Red Team, adversarial testing, and an external assessment. The company integrated MAICyber1Flash into its previously announced multi-agent system, MDASH, which coordinates over 100 specialized AI agents for tasks like code scanning and validation. Microsoft claims the combination achieved a score of 96 percent on the CyberGym benchmark, which measures an AI's ability to spot real security flaws in large codebases. This performance is cited as 12 points higher than Anthropic's Mythos model and ahead of offerings from Google and OpenAI.

A key claimed advantage is cost efficiency....