The record
Written from the 1 report below. Nothing here is unsourced.
- Microsoft has launched MAI-Cyber-1-Flash, its first cybersecurity-specific AI model, available only inside MDASH, its vulnerability identification and remediation system, through an Azure AI Foundry private preview.
- Microsoft says the configuration, which pairs the new model with GPT-5.4, scored 95.95% on CyberGym benchmarks at half the cost of its previous best model mix.
- The headline score belongs to the full MDASH system rather than the new model alone, and the 95.95% result had not appeared on CyberGym's public leaderboard as of late July, while the basis for the cost-saving claim is not fully disclosed.
- The launch is the first announced use case for Project Perception, Microsoft's broader defensive security agent system, which enters public preview on August 3.
What to watch next
- Whether the 95.95% CyberGym result is submitted to and listed on the public leaderboard
- Whether Microsoft discloses the cost-comparison details (token use, call volume, task mix) behind the 50% saving claim
- How the model performs beyond vulnerability management as Project Perception expands after its August 3 public preview
Coverage1
1 report
English national1
All filed from India
Named United States · Azure AI Foundry · CRSBench · CVEBench · CyberGym · CyberSecEval4 · ExploitGym · GPT-5.3 Codex · GPT-5.4 · GPT-5.4 mini · MAI-Cyber-1-Flash · MDASH · Microsoft
The 1 report is listed beside the record.
Ask this story
Answers cite the reports above, or say they can't.