Microsoft Builds Multi-Model AI Security Tool to Compete With Anthropic

Microsoft is preparing to release an AI cybersecurity product internally known as Project Perception, according to people familiar with the company's plans. The system uses models from Microsoft, OpenAI, and Anthropic, selecting among them based on the security task being performed.

The product would give customers access to advanced AI security capabilities without requiring them to commit to one model provider. A routing layer would send vulnerability analysis, code review, threat detection, and incident response work to the model that offers the best balance of performance, speed, and cost. Microsoft reportedly plans to introduce the product in July 2026.

The Information reported the details on July 17, 2026.

Project Perception appears aimed partly at Anthropic's growing influence in cybersecurity. Anthropic's most capable models have demonstrated strong abilities in finding software flaws, analyzing malicious code, and assisting with defensive security operations. Those capabilities have attracted government attention but can be expensive for companies to deploy at scale.

Microsoft has a major distribution advantage through Azure, Windows, GitHub, Microsoft 365, Defender, and its enterprise sales organization. By supporting several model families, it can position itself as the platform that manages security workflows regardless of which AI developer leads individual benchmarks.

The move comes as Microsoft addressed a record 570 vulnerabilities in its July Patch Tuesday, using AI-driven vulnerability discovery in its security operations.

The cybersecurity market is also grappling with a new threat: agentic ransomware. Security researchers documented JadePuffer, described as the first real-world case of fully autonomous AI ransomware that can plan and execute extortion workflows, including reconnaissance, credential theft, and lateral movement, without human direction.