Microsoft released MAI Cyber 1.1 Flash on 27 October as a more focused version of the security model behind MDASH, its vulnerability-scanning product. The company reports comparable performance to MAI Cyber 1 Flash on an internal MDASH evaluation suite at 18% lower cost to complete each scan, with the model taking on more of the work itself so the surrounding harness does less. Microsoft also reports that after a customer's first scan on the MAI Cyber 1 Flash profile, nearly nine in ten subsequent scans use the MAI profile rather than the OpenAI-only profile, and that MAI Cyber now accounts for almost half of all MDASH scans, across thousands of scans from hundreds of customers. The model is available only to verified MDASH customers, under usage policies and ongoing monitoring. The post publishes no external benchmark results and does not break out how many real vulnerabilities were found or how many findings were false positives.
Our takeTwo things cannot be checked from outside: the composition of the MDASH suite, and the per-finding precision and recall behind "comparable performance". A cheaper scan is only equivalent if it surfaces the same real vulnerabilities, and fewer tokens usually shifts the precision-recall balance on tasks like patch generation. The adoption statistic measures which profile customers pick, not which finds more bugs. A held-out set of seeded vulnerabilities, with findings and false positives counted separately, would make the 18% claim falsifiable.