Microsoft Unveils AI Cyber Platform It Says Tops Mythos At Half The Cost
By CU Today Staff —
REDMOND, Wash.--Microsoft has unveiled a new artificial intelligence-powered cybersecurity platform that combines specialized security agents with a new cyber-focused AI model the company says outperforms competing models, including Mythos 5, while costing roughly half as much, Bank Info Security reported.
The launch is designed to give organizations a single platform for continuous AI-driven security monitoring and response without requiring them to build their own AI agents or rely on third-party cyber models, according to Microsoft.
Project Perception combines Microsoft's multi-model security agent harness (MDASH) with a new cybersecurity model, MAI-Cyber-1-Flash, allowing AI agents to work together to identify vulnerabilities, simulate attacks, investigate threats and recommend or implement fixes across identities, endpoints, applications, cloud environments and AI systems, BankInfoSecurity explained.
"The security equation is much more complicated than it used to be," Hayete Gallot, executive vice president of Microsoft Security, said during a briefing. "Our customers have to make a choice between innovation and safety, and that is not acceptable."
Gallot said the platform uses three types of AI agents: "red" agents that identify attack paths, "blue" agents that investigate and prioritize threats, and "green" agents that harden environments and remediate weaknesses. Rather than relying on a single large language model, Perception automatically selects the best AI model for each task from Microsoft's multi-model ecosystem, including Microsoft-developed models and third-party models hosted through its AI platform, BankInfoSecurity explained.
Microsoft also introduced MAI-Cyber-1-Flash, a security-focused foundation model trained on decades of Microsoft's security data. Integrated with MDASH alongside OpenAI's GPT-5.4, Microsoft said the combined system scored 95.95% on the CyberGym evaluation, outperforming GPT-5.5 Cyber (85.6%) and Mythos 5 (83.8%). The company said the model is optimized for code-heavy cybersecurity tasks and automated threat analysis.
Microsoft AI CEO Mustafa Suleyman said the new model not only delivers stronger cybersecurity performance than competing offerings but also costs about half as much, although the company did not disclose pricing. Microsoft said Project Perception is intended to move enterprise security toward a more automated, AI-driven model capable of operating at the speed and scale needed to counter increasingly sophisticated cyberattacks, BankInfoSecurity noted.
Originally reported by CU Today.