Only relevant to security teams on eligible Microsoft Defender XDR plans, and even they need a place in the private preview.
Kenning's take
What it does
MAI-Cyber-1-Flash is Microsoft AI's first cybersecurity model, a compact code-focused model from the MAI-Thinking-1 line. It runs inside Codename MDASH, Microsoft's multi-agent system that scans source code for deep vulnerabilities, handling most tasks and passing the hardest ones to larger models. Microsoft says the combination scores 96% on the CyberGym benchmark at half the cost of its previous setup; MDASH is in private preview for customers with an eligible Microsoft Defender XDR plan.
- Built to find vulnerabilities in large codebases
- Handles about 90% of MDASH tasks, larger models take the rest
- 96% on CyberGym as part of MDASH (Microsoft figure)
- Sandboxed runs with tenant isolation and audit logs
- Used in Microsoft's Perception agentic security system
Who it suits. Enterprise security teams using Microsoft Defender who want AI scanning of their source code.
Key facts
- Category
- AI models
- Made by
- Microsoft AI
- Based in
- United States
- Released
- 13 August 2026
- How to get it
- only inside Codename MDASH, which is in private preview for customers with an eligible Microsoft Defender XDR plan; not sold as a standalone model
- Works on
- Microsoft Defender portal, Defender CLI (Windows, macOS, Linux)
FAQ
When was MAI-Cyber-1-Flash announced?
Microsoft announced it on 13 August 2026.
How can I use MAI-Cyber-1-Flash?
Only through Codename MDASH, which is in private preview for organisations with an eligible Microsoft Defender XDR plan.
What does MDASH scan?
Source code, with particular depth in C, C++, Java and C#. It does not scan compiled binaries.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools

ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.

Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.

Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.