← Back to blog

Microsoft Debuts AI Model for Cybersecurity That Cuts Costs in Half

Microsoft introduces MAI-Cyber-1-Flash within its MDASH platform, achieving high accuracy in vulnerability detection while reducing operational costs. The new model outperforms existing configurations at a fraction of the expense.

TL;DR

  • Microsoft launches MAI-Cyber-1-Flash, its first dedicated cybersecurity AI model.
  • Integrated into MDASH, the model achieved a 95.95% score on CyberGym benchmarks.
  • The solution reduces cost by 50% compared to previous MDASH setups.
  • Leverages a combination of proprietary and GPT-based models for enhanced performance.
  • Access is currently restricted to approved users.

Microsoft has unveiled its latest advancement in cybersecurity automation with the introduction of MAI-Cyber-1-Flash, an AI model designed specifically for identifying and remediating vulnerabilities. Integrated into Microsoft’s MDASH framework, this model demonstrates significant improvements in both performance and cost-efficiency.

In internal testing, MAI-Cyber-1-Flash achieved a 95.95% success rate on the CyberGym benchmark suite. Notably, it delivers these results at approximately half the cost of Microsoft's prior best-performing MDASH configuration, which relied on multiple GPT variants including GPT-5.4, GPT-5.4 mini, and GPT-5.3 Codex.

Performance and Efficiency Gains

  • MAI-Cyber-1-Flash scored 95.95% on standardized CyberGym tests.
  • Costs are reduced by 50% versus previous MDASH model combinations.
  • The model leverages both custom and generative AI technologies like GPT-5.4.

Strategic Implications for Security Teams

  • Improved accuracy can lead to faster incident response times.
  • Lower operational costs make advanced AI tools more accessible to organizations.
  • Limited access suggests Microsoft is piloting the technology with select partners before broader release.

Sources

Sources

Security email updates

One digest email when we publish new security articles (TL;DR plus links to read more). Unsubscribe anytime from the message footer. See our Privacy Policy.