Microsoft touts cost-saving AI model for cybersecurity - CNBC

What the report says

CNBC reported that Microsoft on Monday promoted MAI-Cyber-1-Flash, its first generative AI model designed to identify risky sections of source code, as part of a renewed push in cybersecurity. The company said that, when paired with OpenAI’s general-purpose GPT-5.4, the model outperformed Anthropic’s Mythos 5, Google’s 3.5 Flash Cyber and OpenAI’s GPT-5.5 Cyber on the CyberGym benchmark. Microsoft AI CEO Mustafa Suleyman described the system as offering leading performance at half the cost during a San Francisco event, according to CNBC.

The model is expected to be used in Project Perception, a set of AI agents aimed at finding and helping fix software vulnerabilities. CNBC said the tools are scheduled to enter public preview on Aug. 3, citing a blog post from Hayete Gallot, who returned to Microsoft in February as executive vice president of security after a leadership change in the unit. Project Perception can recommend and carry out code changes with permission and can connect to non-Microsoft products.

The announcement matters because AI is increasingly being used on both sides of cybersecurity. CNBC noted that generative models can help attackers move faster against newly disclosed flaws, while companies such as Anthropic and OpenAI have also released models for defensive security work. Microsoft is positioning its specialized model as part of a broader strategy to improve cost efficiency while continuing its major partnership with OpenAI.

CNBC also reported that Microsoft’s cybersecurity revenue last disclosed in 2023 was above $20 billion annually. The company said MAI-Cyber-1-Flash is not currently offered as a standalone model, and is paired with GPT-5.4 in a system called MDASH to locate exploitable bugs.