At a small event in San Francisco, Microsoft unveiled its first dedicated cybersecurity model, MAI-Cyber-1-Flash, and launched its new AI security platform, Perception. Both products are geared towards enterprise security teams, focusing on vulnerability discovery, remediation, and automation of security processes.
Model access vulnerability detection
Microsoft states that MAI-Cyber-1-Flash is specifically designed to find challenging vulnerabilities in complex codebases and has been integrated into the company's MDASH system for software vulnerability identification and remediation. According to Microsoft, the model outperforms some competing products in both performance and cost-effectiveness in a commonly used industry AI cybersecurity benchmark.
Microsoft's AI chief, Mustafa Suleyman, stated that this suite is ready for immediate deployment in production environments. Microsoft specifically mentioned Anthropic, Google, and OpenAI during the launch event, indicating that this announcement not only represents a product update but also carries a clear competitive undertone.
Perception deploys multiple agents
Perception is designed as a security platform that can deploy multiple types of intelligent agents to assist or automate various security workflows, including vulnerability identification, prioritization, and remediation. The platform can also integrate with MDASH.
- The red team agent was responsible for simulating potential attack paths.
- The blue team is responsible for discovering and distributing existing vulnerabilities.
- The Green Team is in charge of carrying out the repair work.
Hayete Gallot, a Microsoft security executive, said that as attackers increasingly use AI, enterprise defenders also need to use AI to respond to attacks at a similar speed and scale.
Preview available on November 3rd
Dave Weston, head of Perception, stated that tasks that previously required multiple security roles and hours can now be completed in minutes. This system not only identifies problems and prioritizes them, but also provides detection solutions, remedial configurations, and even directly generates code fixes.
Microsoft says the new security tools will be released in preview on November 3. They will enter a market where competition is already heating up. Earlier this year, Anthropic launched its security platform Mythos; OpenAI also released its own security solution in May through Project Day Break.
