Microsoft has announced the release of its first specialised cybersecurity model, MAI-Cyber-1-Flash, as well as a new AI-powered cybersecurity platform called Perception.
According to the company, MAI-Cyber-1-Flash is designed to identify complex vulnerabilities within large codebases and has been integrated into MDASH, Microsoft’s multi-agent vulnerability identification and remediation system.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
Perception is an agentic cybersecurity platform that deploys multiple teams of AI-powered agents to support and automate security workflows, including detecting and fixing software bugs.
The system also integrates with MDASH and uses specialised agent teams: red teams to simulate attacks and identify potential vulnerabilities, blue teams to detect and triage bugs, and green teams to take remedial actions.
Microsoft stated that MAI-Cyber-1-Flash enables MDASH to operate at 50% of the cost of what it considers leading models.
The new model also achieved a score of 96% on the CyberGym benchmark, reportedly outperforming other models referenced by Microsoft, such as Gemini, Mythos, and GPT.
The company explained that MAI-Cyber-1-Flash processes up to 90% of tasks efficiently, with more demanding challenges being handled by larger models in Microsoft’s portfolio, like GPT-5.4. This arrangement is intended to improve both cost-effectiveness and performance.
Microsoft AI CEO Mustafa Suleyman and Microsoft Security EVP Hayete Gallot, in a blog, said: “MAI-Cyber-1-Flash is a compact, code-heavy security model derived from the MAI-Thinking-1 lineage, which was built from scratch, in-house, on the highest quality data.”
The company says the model was trained on decades of security signals and real-world attack data, which it claims underpin its performance. Both MDASH and MAI-Cyber-1-Flash were developed with safety in mind.
Microsoft said the model has been evaluated by its internal AI Red Team, supported by adversarial testing and independent assessments.
MDASH customers are provided with enterprise-grade controls, including role-based access, data isolation, encryption, audit capabilities, and sandboxed environments.
The Perception platform is scheduled to enter public preview on 3 August.
Microsoft said the platform continuously learns and adapts by connecting signals, context, models, and agents.
Perception is intended to help organisations understand risk, prioritise threats, and apply protections in real-time, while keeping human users in control of decision-making.
Earlier this month, Microsoft and Mistral announced an expanded partnership to support the use of AI models by businesses and organisations in regulated sectors.
