Microsoft is enhancing its Windows security efforts by integrating artificial intelligence to accelerate the detection and prioritization of software vulnerabilities. The company's new multi-model agentic scanning harness, known as MDASH, utilizes multiple AI models, including third-party solutions, to scan the Windows codebase for potential issues.
This AI-driven approach is designed to identify vulnerabilities faster, reduce the time it takes to address them, and scale the discovery process. According to Pavan Davuluri, executive vice president of Windows + Devices, confirmed vulnerabilities then undergo a Windows-specific validation process to eliminate false positives, ensuring that only high-confidence findings are presented to the engineering team.
Davuluri emphasized that the primary goal of these AI tools is to expedite the process from vulnerability discovery to customer protection. He noted that as AI improves defenders' ability to find issues, users can expect to see a higher volume of security updates. This increased frequency of updates will signal an improvement in the detection and remediation of security flaws.
Despite the growing reliance on AI for detection, Microsoft maintains that human expertise remains crucial for key security decisions and the implementation of fixes. The company aims to use AI to support faster protection, strengthen engineering systems, and provide more actionable guidance to customers.
Microsoft's adoption of AI in cybersecurity is part of a broader trend. Other companies, like Firefox, have also reported significant success using AI for bug hunting. Earlier this year, Firefox's CTO praised an AI model for identifying hundreds of vulnerabilities in the browser. Last year, an AI even topped a leaderboard for system vulnerability hunting.
The increasing use of AI in cybersecurity is partly a response to threat actors who are also leveraging AI for malicious purposes. Examples include AI-assisted hacking groups using deepfakes and spoofed communications for social engineering scams. However, AI models themselves can be manipulated, with researchers finding ways to bypass their safety guardrails. The cybersecurity landscape is evolving into a competitive race where both defenders and attackers are increasingly utilizing machine learning.