Scan costs cut by 18%
Microsoft on Oct. 27 (Oct. 28 Japan time) released "MAI-Cyber-1.1-Flash," an improved version of its AI model for finding software vulnerabilities. It built the model into "MDASH," its vulnerability scanning system, improving the accuracy of the flaws it finds and reducing the tokens it uses, the company said. @aitrackerbot, an X account that tracks new model releases, was among the first to report it.
According to Microsoft, internal evaluations showed it cut the cost per scan by 18% while performing at about the same level as the previous version. This is not a price cut for customers. It is available only to approved MDASH customers.
The model handles half of scans
Mustafa Suleyman, who leads Microsoft's AI division, posted on X that the share of MDASH scans using this model doubled to 50% in 30 days. The company's announcement also said the model handles nearly half of MDASH scans. It did not give the original share.
MDASH is a system in which more than 100 agents use multiple models to find, verify and fix vulnerabilities. The first version, "MAI-Cyber-1-Flash," was released on Aug. 13. Microsoft said it halved costs by having the model handle 90% of the work and passing the remaining, difficult 10% to OpenAI's large models.
Companies compete with AI for attack and defense
AI that finds vulnerabilities is an area of focus for many companies. Anthropic provides "Claude Mythos," aimed at cyber defense, to some organizations, and Cognition revamped "Devin Security," which uses it. Microsoft claims that the original MDASH's results surpassed Mythos.
