Artificial Intelligence & Machine Learning,
Next-Generation Technologies & Secure Development,
The Future of AI & Cybersecurity
Project Perception Combines AI Agents With New MAI-Cyber-1-Flash Model
On July 27, 2026, Microsoft made a significant announcement regarding its cybersecurity offerings, unveiling a sophisticated new security agent stack and a cost-effective cyber-capable model. The company claims that this new model can surpass the capabilities of existing technologies such as Mythos, while being priced at just half the cost.
This development opens doors for Microsoft’s customers to access a consolidated platform that facilitates continuous artificial intelligence-driven security surveillance and protective measures. The platform eliminates the need for organizations to independently develop their own security agents or engage various AI vendors to utilize top-tier cybersecurity models.
Referred to as Project Perception, the integrated agent platform aims to cultivate a deep, ongoing awareness of the security environment surrounding an organization. It achieves this by combining specialized agents and models, integrating Microsoft’s multi-model security agent known as MDASH with the innovative MAI-Cyber-1-Flash cybersecurity model.
Hayete Gallot, the executive vice president of Microsoft Security, emphasized the pressing challenges posed to enterprises by contemporary attackers, asserting the necessity for organizations to bolster their defensive capabilities to match the speed and intensity of these threats. “The security equation is much more complicated than it used to be,” stated Gallot, highlighting the unacceptable reality that companies are often forced to choose between innovation and safety.
Gallot elaborated that Project Perception is a pioneering kind of cybersecurity stack that aspires to facilitate a more automated approach to security management. The platform employs three distinct types of agents: red agents tasked with identifying attack vectors, blue agents responsible for defense and risk assessment, and green agents dedicated to rectifying vulnerabilities and fortifying security postures.
Users can interact with Perception by simply posing questions—such as evaluating vulnerabilities related to specific types of attacks. The platform’s agents then collaborate, selecting the most appropriate agents required to fulfill the task. Initially, an agent will delve into the company’s codebase to pinpoint possible vulnerabilities, then pass this information to a subsequent agent to construct and simulate the attack path. Finally, another agent will produce customized vulnerability alerts while concurrently addressing necessary patches.
“We see across identities, endpoints, applications, data, clouds, and AI systems, providing broad visibility across the digital estate,” commented Gallot in a blog post. “Equally important, we can help customers take action across those environments. Combined with decades of security research, threat intelligence, and real-world operational experience defending organizations, these capabilities shape how Project Perception reasons, prioritizes, and responds.”
Project Perception adopts a multi-model approach; its agents are not restricted to specific models, whether from Microsoft, OpenAI, or other large language models hosted within Microsoft’s model garden. This flexibility enables the agents to choose the model that is best suited for the task at hand.
Despite the platform’s access to premier AI models, Microsoft continues to develop its own foundation models. Integral to Project Perception is the introduction of MAI-Cyber-1-Flash, a compact, code-heavy model grounded in Microsoft’s MAI-Thinking-1 large language model (LLM). This new cyber model has been trained on the extensive security data that Microsoft has accumulated over the decades and is designed to integrate seamlessly within the MDASH framework.
An official blog post unveiling MAI-Cyber-1-Flash revealed that it, combined with OpenAI’s GPT-5.4, outperformed several existing cyber-capable models, such as Mythos 5, GPT-5.5 Cyber, and GPT-5.6 Sol during the CyberGym evaluation. The scores highlighted the efficacy of the MDASH harness, which achieved an impressive 95.95%, compared to Mythos 5’s 83.8% and GPT-5.5 Cyber’s 85.6%.
Mustafa Suleyman, CEO of Microsoft AI, noted that the new model provides not only superior performance compared to its counterparts but also does so at half the cost. However, the company has yet to disclose any specific pricing information pertaining to the model. As Microsoft pioneers advancements in artificial intelligence and cybersecurity, Project Perception marks a substantial step forward in addressing current security challenges while shaping the future of these technologies.