- Microsoft launched MAI-Cyber-1-Flash, an AI model designed for cybersecurity tasks
- Project Perception uses AI agents to detect vulnerabilities, assess risks, and automate fixes
- MDASH platform, powered by MAI-Cyber-1-Flash, will enter public preview on August 3
Microsoft has introduced MAI-Cyber-1-Flash, its first proprietary AI model built for cybersecurity, alongside Project Perception, an AI-powered framework that helps detect software vulnerabilities, assess cyber risks and automate fixes. The rollout will begin with software vulnerability management.
“We really do have a pretty significant data and harness and expertise moat, and that is enabling us to train models which are faster, better, cheaper, and I think this is genuinely the tip of the iceberg," Microsoft AI CEO Mustafa Suleyman told VentureBeat.
Project Perception integrates Microsoft's MAI-Cyber-1-Flash model with the MDASH platform first introduced in May. Built into Microsoft Defender, the service is set to enter public preview on Aug. 3. Its availability will be gradually extended to all products within the Microsoft Security ecosystem.
ALSO READ | iPhone 17 To Get Costlier By Rs 12,000 From August? Check What Tipster Claims
Microsoft's latest announcement comes as more tech companies race to build AI tools for cybersecurity. Earlier this year, Anthropic introduced Mythos to a select group of partners under its Glasswing programme, while OpenAI launched its Daybreak security offering in May, according to TechCrunch.
The company said MAI-Cyber-1-Flash is a lightweight, code-focused version of its MAI-Thinking-1 family, trained on Microsoft's extensive database of security exploits and past fixes. The model now handles nearly 90% of the workload within MDASH, Microsoft's multi-model vulnerability scanning system unveiled in May. More complex cases are passed on to OpenAI's GPT-5.4. According to Microsoft, this setup has reduced MDASH's operating costs by roughly 50%.
Microsoft also claimed that MDASH, powered by MAI-Cyber-1-Flash, scored 95.95% on the CyberGym benchmark, which includes 1,507 vulnerability reproduction tests. By comparison, the company said Anthropic's Mythos achieved a score of around 84%. When MDASH was first introduced in May, it had scored 88.45% on the same benchmark.
The earlier version of MDASH also helped uncover 16 previously unknown vulnerabilities in Windows networking and authentication components. Microsoft said four of them were critical remote code execution flaws, which were patched as part of its May Patch Tuesday security update.
Microsoft also announced Project Perception, an agentic cybersecurity platform that orchestrates specialised AI agents for different security tasks. The system assigns "red team" agents to identify potential attack paths, "blue team" agents to assess and prioritise risks, and "green team" agents to strengthen defences and carry out remediation. The public preview is scheduled to begin on Aug. 3.
ALSO READ | AI Will Help India Shed Its 'World's Back Office' Tag: 1Point1 Solutions CMD | Interview
“Project Perception brings together signals, context, models and specialised agents into a continuously learning system of defence. It can reason, prioritise and act at machine speed while keeping humans firmly in control and empowering them with powerful new workflows,” Microsoft wrote on its blog.
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.