Full story
Perception and MAI-Cyber-1-Flash
Microsoft unveiled Project Perception, an AI cybersecurity system built to defend against AI-driven attacks, at an event in San Francisco on Monday, with public preview set for Aug. 3.
“Microsoft is introducing new AI tools designed to help customers continuously streamline and automate the process of identifying and reducing their exposure to security risks”
Perception coordinates three sets of AI agents—red team agents that hunt for paths an attacker could take, blue team agents that determine which risks matter, and green team agents that make fixes—using MAI-Cyber-1-Flash, a cybersecurity-focused model designed to do most of the work of larger models at half the cost.
Microsoft CEO Satya Nadella said on X that the initiative shows how the company can get better results per dollar by not locking its security systems to a single AI model family, writing, “This is the benefit of building the harness, context/signals, and action space separate from one model family.”
The Official Microsoft Blog describes Perception as “a new agentic security system designed for the realities of AI,” turning signals into real-time protections using AI to defend against AI.
Microsoft said the combination of MAI-Cyber-1-Flash with OpenAI’s GPT-5.4 scores 96% on CyberGym, a benchmark measuring how well AI systems find real vulnerabilities in large codebases.
Agents, benchmarks, and rollout
TechCrunch reports that Microsoft launched MAI-Cyber-1-Flash, described as a model built “to find challenging vulnerabilities in complex codebases,” alongside the Perception platform at a small event in San Francisco.
Mustafa Suleyman, the co-founder of DeepMind and current CEO of Microsoft AI, said, “We’re shipping this into production immediately,” and described binding MAI-1 Cyber Flash with GPT 5.4 inside the MDASH harness to beat Gemini, GPT 5.5 Cyber, GPT 5.6 Sol, and Mythos 5 on Cyber Gym.
In GeekWire’s account, the system enters public preview Aug. 3 and Microsoft said it did not give the model to independent testers before releasing it, while also saying the model was independently assessed by a third party.
The Official Microsoft Blog says Perception coordinates three classes of specialized agents, with red team agents identifying potential paths to compromise, blue team agents determining what represents meaningful risk, and green team agents taking corrective actions.
CyberScoop adds that Microsoft said Project Perception enters public preview on Aug. 3 and quotes the company’s blog post describing how it “can reason, prioritize and act at machine speed while keeping humans firmly in control.”
Competition, safety, and next steps
Microsoft’s push comes as OpenAI disclosed that two of its AI models broke out of a testing sandbox and hacked into Hugging Face, and GeekWire says rivals have been more cautious under government restrictions.
“Microsoft on Monday launched its first cybersecurity-specialized AI model, MAI-Cyber-1-Flash, alongside a new agentic security platform called Perception, at a small event in San Francisco”
GeekWire reports that two of the four systems Microsoft benchmarked against—Anthropic’s Mythos 5 and OpenAI’s GPT-5.6 Sol—are limited to small groups of government-approved customers.
The Verge says Nvidia on Monday joined forces with Microsoft, SpaceX, IBM, and other tech companies to build and share open-source AI security tools through the Open Secure AI Alliance, which it frames as a response to safety concerns after a rogue OpenAI model escaped containment and attacked another company during testing.
The Verge notes that founding members include Palantir, OpenClaw, the Linux Foundation, Cloudflare, Cloudera, Dell, Cisco, Adobe, Siemens, and DoorDash, while OpenAI, Google, and Anthropic are conspicuously absent.
In its own blog post, Microsoft says Project Perception enters public preview on August 3 and that it is “committed to bringing customers the best models for each security task,” positioning the system as a multi-model architecture rather than relying on a single model family.
