July 29, 2026, (Inside AI) — Microsoft launched two AI cybersecurity products on July 27: MAI-Cyber-1-Flash, a specialized model for finding code vulnerabilities, and Perception, an agentic platform that orchestrates AI agent teams to identify, prioritize, and fix security flaws. The launch marks a direct entry into a market where rivals Anthropic and OpenAI already operate with their own programs.
Perception deploys three types of AI agent teams: red teams simulate attacks, blue teams detect and triage bugs, and green teams generate and implement code fixes. The workflow compresses a process that once took hours across multiple roles into minutes, according to Dave Weston, lead engineer for Perception.
"We're very very excited to announce our results," said Mustafa Suleyman, CEO of Microsoft AI. "We have MAI-1 Cyber Flash binded with GPT 5.4 inside of the MDASH harness -- which beats out Gemini, GPT 5.5 Cyber, GPT 5.6 Sol, and Mythos 5 on Cyber Gym, which is the primary benchmark that we all use. The golden benchmark. We're shipping this into production immediately."
The model automates vulnerability hunting by understanding code architecture and threat patterns. It targets a long-standing problem: security teams spend exhaustive hours manually searching for exploitable gaps, often during off-hours. MAI-Cyber-1-Flash acts as a tireless partner, freeing analysts for strategic work.
Microsoft claims MAI-Cyber-1-Flash outperforms competitors on the Cyber Gym benchmark, beating offerings from OpenAI, Google's Gemini, and Anthropic's Mythos 5. The company's benchmarking shows it leading in vulnerability detection accuracy, though independent verification is pending.
Agentic Architecture Redefines Security Operations
Perception's three-team structure simulates real attacker behavior. Red teams model threat actors to surface likely exploit targets. Blue teams then prioritize bugs based on risk, not human guesswork. Green teams execute fixes, closing gaps automatically. This shifts defenders from reactive firefighting to proactive strategy.
The platform's efficiency addresses an inverted threat landscape. Hackers already use AI to automate attacks, leaving manual defenders behind. Smaller organizations benefit most: a startup with two engineers can now achieve coverage that previously required ten, leveling the playing field against well-resourced adversaries.
Perception enters public preview on November 3, competing directly with Anthropic's Glasswing program and OpenAI's Daybreak initiative. Microsoft's integration of its model with GPT 5.4 in the MDASH harness suggests a hybrid approach that could set a new standard for AI-driven security operations.
