Microsoft is preparing to launch Project Perception, an AI-powered security platform designed to lower the cost of continuous vulnerability discovery. The move intensifies competition with agentic AI security tools and specialized providers like Intruder, as enterprises race to automate pentesting against increasingly sophisticated, machine-speed cyber threats.
Microsoft’s Project Perception and the Shift Toward Multi-Model Security
Microsoft is set to debut Project Perception before the end of July 2026, a platform engineered to scan enterprise codebases for exploitable vulnerabilities. By reserving high-cost frontier models for critical analysis and using cheaper, distilled models for routine scans, Microsoft aims to undercut the price of current market leaders.
The platform represents a strategic pivot for Microsoft’s security division under executive vice president Hayete Gallot, who returned to the company from Google Cloud in February 2026. Project Perception is built on the company’s internal Multi-model Agentic Scanning Harness (MDASH), a system developed by the Autonomous Code Security team to automate deep-level security analysis.
Intruder’s Web Application Pentesting and the Cost of Accessibility
While Microsoft targets the enterprise, Intruder has launched its own AI-driven pentesting product aimed at small and medium-sized businesses. The service, which integrates with GitHub and GitLab, allows organizations to run white-box tests that provide visibility into application source code. Intruder claims its automated testing can begin in minutes and conclude in hours, a sharp contrast to traditional engagements that often require weeks of scheduling.
Pricing for the service starts at USD $3,500 per test, which the company frames as 25% or less of the cost of a traditional manual pentest. Intruder founder Chris Wallis emphasized that the economics of legacy security testing are failing to match modern development cycles.
The Competitive Gap Between Claude Mythos and Public Access
The market for AI-driven security is currently defined by a significant access divide. On April 7, 2026, Anthropic launched Claude Mythos Preview but restricted it to a controlled-access consortium
of 50 organizations known as Project Glasswing. Partners, including companies like AWS, Cisco, and Microsoft, received $100 million in usage credits to test the model.
The results of the Glasswing program demonstrated the high-stakes nature of the technology. Anthropic used Mythos to scan over 1,000 open-source projects, flagging 6,202 high- or critical-severity vulnerabilities. Independent firms confirmed 90.6% of those findings to be valid. However, because Anthropic has not released Mythos for general public use—citing warnings to US officials that the model could facilitate large-scale cyberattacks—most enterprises remain locked out of the most capable tools.
Microsoft is positioning Project Perception to fill this void. While Anthropic has faced regulatory hurdles, including export restrictions on its Fable 5 model, Microsoft’s global data-center footprint—spanning more than 60 regions—gives it a structural advantage in serving international markets that currently lack access to Mythos-class tools.
Security Leaders Face a 68% Coverage Gap
The push for agentic AI in security stems from a persistent, industry-wide coverage gap. According to data cited by Synack, 95% of security leaders at large US enterprises prioritize pentesting, yet the average organization tests only 32% of its attack surface. This leaves a 68% blind spot that attackers are increasingly exploiting using autonomous AI agents.
The urgency is felt by end-users like Yembo, an AI platform that has turned to Intruder’s automated testing to supplement its human-led security efforts. While Yembo continues to leverage human pentesters, annual assessments alone leave dangerous windows of exposure,
said CTO Zach Rattner. Intruder’s AI pentesting bridges that gap by delivering human-grade depth at machine speed to keep our platform permanently hardened.
As these tools move from internal testing to widespread enterprise deployment, the industry is shifting away from the photograph of a building
model of annual pentesting toward a continuous, machine-speed defense. The competition between Microsoft’s multi-model orchestration and specialized platforms like Intruder will likely hinge on which provider can best balance the need for autonomous speed with the rigorous human validation required for high-assurance environments.
Discover more from Archyworldys
Subscribe to get the latest posts sent to your email.