News
Microsoft released its first autonomous large-scale security model, MAI-Cyber-1-Flash, which outperformed its competitors in MDASH system benchmark tests. Microsoft today announced two major developments in the field of network security: the launch of its first self-developed AI model specifically designed for network security, MAI-Cyber-1-Flash, and a new model named MAI-Cyber-1-Flash.
2 min read
Source: Telegram AI频道
Microsoft released its first independent large-scale security model, MAI-Cyber-1-Flash, which outperformed its competitors in MDASH system benchmark tests. Microsoft today announced two major developments in the field of network security: the launch of its first self-developed AI model specifically designed for network security, MAI-Cyber-1-Flash, and an intelligent AI security system called Project Perception. As a compact and code-focused security model, MAI-Cyber-1-Flash is developed based on the MAI-Thinking-1 series architecture. Microsoft integrated it into the previously released evolved version of the multi-model agent security system MDASH. With the new architecture, MAI-Cyber-1-Flash can efficiently carry up to 90% of the workload in the MDASH system, while the remaining 10% of complex tasks requiring deep reasoning are handled by OpenAI's GPT-5.4. After testing, this new combination achieved a success rate of 95.95% in the CyberGym benchmark, which was significantly improved compared to the previous MDASH's 88.45%, and defeated GPT-5.5 Cyber (85.6%) and Anthropic's Mythos 5 (83.8%). In addition to performance improvement, this combination also significantly reduces operating costs, saving nearly 50% compared to the original MDASH configuration using GPT-5.4, GPT-5.4 Mini and GPT-5.3 Codex. At the same time, the Project Perception system announced by Microsoft is designed to provide continuous dynamic protection for enterprises. The system consists of three professional agents: red team, blue team and NVIDIA. The red team agent is responsible for simulating attacks and identifying potential attack paths, the blue team is responsible for troubleshooting and assessing risk priorities, and NVIDIA focuses on applying patches and consolidating defenses. iframe (www.youtube.com) Project Perception's public beta will be launched on August 3, when it will be available to users as part of Microsoft Defender and adopt a pay-as-you-go flexible payment model. In the future, Microsoft also plans to further expand the application scope of MAI-Cyber-1-Flash to more security workflows beyond software vulnerability management. via cnBeta.COM - Chinese industry information station (author: source: cnBeta.COM)