On Monday, Microsoft announced a new artificial intelligence model, MAI-Cyber-1-Flash, designed to identify cybersecurity vulnerabilities, representing a strategic revitalization of its cybersecurity division under the leadership of Hayete Gallot.
This model, when used alongside OpenAI's GPT-5.4, reportedly outperforms competitors like Anthropic's Mythos 5 and Google's 3.5 Flash Cyber on the CyberGym benchmark, with CEO Mustafa Suleyman claiming it offers 'world-leading performance at 50% of the cost.' The model is part of Project Perception, which will enter public preview on August 3.
Gallot, who rejoined Microsoft in February, aims to enhance the cybersecurity workforce by making it easier for companies to staff their security operating centers (SOCs). Despite a 19% decline in Microsoft shares this year, analysts, including Karl Keirstead, recommend buying the stock, citing concerns over the company's exposure to OpenAI and competition from open-source AI models.
Microsoft has not disclosed recent revenue figures for its cybersecurity business, which exceeded $20 billion in 2023. The company continues to innovate in this space, having previously launched the Security Copilot assistant and integrating AI capabilities into its productivity tools.
The potential for further improvements in the new model's performance is significant, as Suleyman noted that less than 1% of their unique data set has been utilized so far