Microsoft Unveils First Internally Developed Cyber Model
New tools aim to find, rank and patch vulnerabilities faster while lowering enterprise security costs.
Topics
News
- Fired OpenAI Researchers Warn Dismissals Could Chill Safety Work
- ARTEX Developer Closes AI Hacking Tool After Bank Attacks
- AI Redraws Global Race for Quant Talent as Coding Tools Lower Entry Barriers
- AI Dispatch | India Draws Billions Into AI
- AI Bots Now Make Up Half of Traffic to US Fed’s FRED Database
- Musk Blames ‘Oligarchs’ For Delay in Starlink’s India launch
Image Credit- Chetan Jha/ MIT Sloan Management Review India
Microsoft Corp. on Monday unveiled MAI-Cyber-1-Flash, its first cybersecurity model built in-house, at a security event in San Francisco.
The model was launched alongside Project Perception, an agentic system that coordinates specialized AI agents across vulnerability detection, risk assessment and patching.
Project Perception initially uses three agents: the red agent searches for software weaknesses, blue assesses which findings pose the greatest risk, and the green agent develops and applies fixes.
Microsoft said the agents share security information and perform tasks similar to those handled by human security teams. The system is built on MDASH, the company’s framework of more than 100 agents used to identify, validate and remediate vulnerabilities.
MAI-Cyber-1-Flash is a smaller, code-focused model trained specifically for cybersecurity work. Microsoft said it can handle about 95% of the vulnerability-analysis tasks performed by MDASH.
More complex work is passed on to OpenAI’s GPT-5.4. Microsoft said this approach reduces costs by using the larger model only when additional computing power is required.
The combined system scored 95.95% on CyberGym, a benchmark that tests whether AI models can produce working proof-of-concept exploits for known vulnerabilities, according to Microsoft.
The company said competing models, including Anthropic’s Mythos and OpenAI’s GPT-5.5-Cyber, scored about 83%. However, the benchmark results and comparisons have not been independently verified.
Microsoft said MAI-Cyber-1-Flash would cost roughly half as much as systems that rely mainly on large general-purpose models.
The company will make the model available through Azure AI Foundry rather than releasing it publicly. Access will use Microsoft’s customer screening and computing-provisioning processes.
Microsoft said the system includes role-based access controls, tenant separation, encryption, audit records and isolated execution environments without internet access.
The launches come as technology companies develop specialized cybersecurity models to help defenders respond to attackers using AI to find and exploit software weaknesses more quickly.
