Microsoft Intros New Agentic AI Security Multi-Model Defense System — Campus Technology

Microsoft Intros New Agentic AI Security Multi-Model Defense System

A brand new multi-model agentic AI safety system constructed by Microsoft’s Autonomous Code Security staff helped researchers discover 16 new vulnerabilities throughout the Home windows networking and authentication stack, the corporate anounced in a current safety weblog submit.

The Redmond-based firm is hedging its future safety operations facilities on using coordinated AI brokers to help its staff finishing up standard safety operations. 

Microsoft stated its inside system, codenamed MDASH (Microsoft Security multi-model agentic scanning harness), helped researchers uncover 16 beforehand unknown vulnerabilities throughout Home windows networking and authentication elements, together with 4 essential distant code execution flaws.

Not like conventional AI safety instruments that depend on a single mannequin, the software program big stated, “MDASH coordinates greater than 100 specialised AI brokers operating throughout a number of frontiers and distilled fashions.”

Microsoft stated the system achieved industry-leading benchmark outcomes, together with an 88.45 % rating on the CyberGym benchmark, which covers greater than 1,500 real-world vulnerabilities.

MDASH affords perception into Microsoft’s broader push towards what it calls “agentic safety,” the place autonomous AI methods more and more help — and in some circumstances, automate — menace detection, investigation, and remediation.

Researchers from Crew Atlanta, the group that received $20 million in DARPA’s AI Cyber Problem, helped create MDASH. Microsoft is positioning analysis into MDASH as a part of an overarching method to show AI-powered vulnerability analysis into scalable, production-grade safety engineering.

Taesoo Kim, Microsoft’s vp of agentic safety, famous that the system is designed to research code autonomously, debate exploitability, validate findings and generate proof-of-concept exploits.

This train illustrated Microsoft’s positioning of AI not simply as a productiveness instrument for defenders, however as a core operational layer for figuring out and mitigating vulnerabilities earlier than attackers can exploit them.

For extra data, learn the Microsoft weblog.