Microsoft has introduced its first cybersecurity-specific artificial intelligence model, MAI-Cyber-1-Flash, operating inside its multi-model vulnerability identification and remediation harness known as MDASH according to [Thehackernews](https://thehackernews.com/2026/07/microsoft-says-new-cybersecurity-ai.html). Revealed at an event in San Francisco, the system pairs the new model with OpenAI’s GPT-5.4 [according to](https://www.cnbc.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html) [CNBC](https://www.cnbc.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html). Mustafa Suleyman, CEO of Microsoft AI, stated that the configuration achieves world-leading performance at half the cost of previous setups [according to](https://www.cnbc.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html) [CNBC](https://www.cnbc.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html).
Microsoft Launches MAI-Cyber-1-Flash and MDASH Harness
According to Microsoft’s model card, MAI-Cyber-1-Flash is a sparse mixture-of-experts transformer featuring 137 billion total parameters, 5 billion active parameters, and a 256,000-token context window [according to](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html) [Thehackernews](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html). It was developed from a MAI-Code-1-Flash fine-tune, originating from a MAI-Thinking-1 mid-training checkpoint [according to](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html) [Thehackernews](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html). The architecture is designed to route tasks efficiently, with MAI-Cyber-1-Flash handling up to 90% of MDASH tasks while GPT-5.4 is reserved for the hardest 10% [according to](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html) [Thehackernews](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html).
Benchmark Performance and Cost Savings
The configuration scored 95.95% on the CyberGym benchmark according to [Thehackernews](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html), while Microsoft’s published benchmarks cite a 96% score compared to Anthropic’s Mythos 5 at 84% [according to](https://www.cnet.com/tech/services-and-software/microsoft-says-its-new-cybersecurity-ai-beats-industry-leaders-at-half-the-cost/) [cnet.com](https://www.cnet.com/tech/services-and-software/microsoft-says-its-new-cybersecurity-ai-beats-industry-leaders-at-half-the-cost/). Microsoft reported that this performance outperforms Anthropic’s Mythos, Google Gemini, OpenAI GPT-5.5 Cyber, and GPT-5.6 Sol [according to](https://arstechnica.com/security/2026/07/27/microsoft-unveils-ai-security-tools-it-says-outperform-competing-platforms/) [Ars Technica](https://arstechnica.com/security/2026/07/27/microsoft-unveils-ai-security-tools-it-says-outperform-competing-platforms/).
The model card notes that the evaluated configuration replaced 80% of MDASH’s existing models, raising reported CyberGym results from 88.4% to 95.95% [according to](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html) [Thehackernews](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html). Microsoft states the setup delivers comparable performance at 50% of the cost of its previous best MDASH combination of GPT-5.4, GPT-5.4 mini, and GPT-5.3 Codex [according to](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html) [Thehackernews](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html). Pricing is consumption-based, measured by security compute units (SCUs) consumed as AI agents run scenarios [according to](https://www.cnet.com/tech/services-and-software/microsoft-says-its-new-cybersecurity-ai-beats-industry-leaders-at-half-the-cost/) [cnet.com](https://www.cnet.com/tech/services-and-software/microsoft-says-its-new-cybersecurity-ai-beats-industry-leaders-at-half-the-cost/).
Project Perception Integration and Availability
MAI-Cyber-1-Flash and MDASH form the core of Project Perception, a collection of specialized AI agents that execute red-, blue-, and green-team functions to discover vulnerabilities, investigate risks, and enact corrective actions [according to](https://arstechnica.com/security/2026/07/27/microsoft-unveils-ai-security-tools-it-says-outperform-competing-platforms/) [Ars Technica](https://arstechnica.com/security/2026/07/27/microsoft-unveils-ai-security-tools-it-says-outperform-competing-platforms/). According to Hayete Gallot, Microsoft’s executive vice president of security, Project Perception is scheduled to enter public preview on August 3, built directly into Microsoft Defender before rolling out to other Microsoft Security products [according to](https://www.cnet.com/tech/services-and-software/microsoft-says-its-new-cybersecurity-ai-beats-industry-leaders-at-half-the-cost/) [cnet.com](https://www.cnet.com/tech/services-and-software/microsoft-says-its-new-cybersecurity-ai-beats-industry-leaders-at-half-the-cost/).
Access is currently limited to approved MDASH customers through an Azure AI Foundry private preview [according to](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html) [Thehackernews](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html). Microsoft plans to extend the model beyond software vulnerability work into additional security workflows [according to](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html) [Thehackernews](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html), though model cards warn that generated text and code may be inaccurate or incomplete and require review prior to consequential deployment [according to](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html) [Thehackernews](https://thehackernews.com/2026/07/27/microsoft-touts-cost-saving-ai-model-for-cybersecurity.html).
Keep reading
