Microsoft unveils its first cybersecurity model and the Perception agent platform
Translated by a model; the original is in Russian.
On Monday, at a closed event in San Francisco, Microsoft unveiled its first cybersecurity-focused model, and alongside it a new platform for defense. With this move, the company steps directly into competition with Anthropic, Google and OpenAI.
The MAI-Cyber-1-Flash model
MAI-Cyber-1-Flash is described as a model built to hunt down complex vulnerabilities in tangled codebases. It powers MDASH, Microsoft's internal tool for detecting and fixing software defects.
According to Mustafa Suleyman, head of Microsoft AI and co-founder of DeepMind, pairing MAI-Cyber-1-Flash with GPT 5.4 inside MDASH outperforms competitors — Gemini, GPT 5.5 Cyber, GPT 5.6 Sol and Mythos 5 — on the Cyber Gym benchmark, which he called the industry's chief yardstick. The company claims its model is significantly more powerful while also being more cost-effective than its counterparts. “We're taking this into production immediately,” Suleyman added.
The Perception platform
The second new release is Perception, a platform that deploys teams of agents to automate a range of security tasks, including finding and fixing bugs. It, too, can integrate with MDASH.
Microsoft's vice president for security, Ayete Gallo, noted that attackers are increasingly turning to AI in their attacks, and presented Perception as a way to “defend against AI with AI” — at the same speed and scale as the attackers.
Perception relies on agentic “red,” “blue” and “green” teams. The red teams model possible attacks, profiling likely adversaries and the vulnerabilities they might exploit. The blue teams handle detecting and triaging existing bugs, while the green teams carry out corrective action.
Dave Weston, the platform's lead engineer, emphasizes the efficiency gains: what once took hours of manual work by a host of narrow specialists — from vulnerability hunters to remediation engineers — now takes minutes. The system not only finds and prioritizes problems but also proposes a ready-made code fix.
A crowded market
Microsoft's new tools will become available in preview on November 3. They will enter an increasingly crowded market: earlier this year Anthropic launched its Mythos platform for a select group of partners under the Glasswing program, and in May OpenAI unveiled its own offering under the Daybreak program.
Source: TechCrunch
Comments
Sign in, to comment.