Microsoft unveiled its first cybersecurity-specific mannequin and new AI cybersecurity platform at a small occasion in San Francisco on Monday, giving it a giant win over the main gamers within the area: Anthropic, Google, and OpenAI.
The corporate describes MAI-Cyber-1-Flash as a mannequin constructed “to seek out tough vulnerabilities in complicated codebases.” This mannequin is constructed to animate MDASH, Microsoft’s harness devoted to figuring out and remediating software program vulnerabilities.
The brand new safety platform is known as Notion and is designed to deploy a staff of brokers to help and automate numerous safety workflows, together with figuring out and remediating bugs. This platform can be built-in with MDASH.
The corporate claims MAI-Cyber-1-Flash is considerably extra highly effective (and cheaper) than competing fashions primarily based on its efficiency on established AI cybersecurity benchmarks.
Mustafa Suleiman, co-founder of DeepMind and present CEO of Microsoft AI, mentioned: “We use MAI-1 Cyber Flash certain to GPT 5.4 throughout the MDASH harness. It outperforms Gemini, GPT 5.5 Cyber, GPT 5.6 Sol, and Cybergym’s Mythos 5. That is the first benchmark all of us use. It is the golden benchmark.”
“We’ll be transport this into manufacturing quickly,” he added.
Hayete Gallot, vp of safety at Microsoft, famous that hackers are more and more utilizing AI in cyberattacks, and mentioned Notion is a method for enterprise defenders to “defend towards AI utilizing AI on the scale and pace of attackers.”
Notion makes use of crimson, blue, and inexperienced groups of brokers. Pink Teaming gives detailed simulations of potential assaults and gives context about potential menace actors and the vulnerabilities they could exploit. The blue staff focuses on detecting and prioritizing current bugs, whereas the inexperienced staff takes “corrective actions” towards these bugs.
Dave Weston, principal engineer at Notion, described the platform as a big effectivity improve for enterprise defenders. “We spent hours of guide effort from a number of professional employees throughout our safety group (appsec hunters, remediation engineers, and so forth.) to resolve this concern, and we had been in a position to repair all of this in minutes. We not solely uncover and prioritize points, we additionally detect, remediate state, and even remediate code.”
Whereas AI has supplied companies with new defensive capabilities, the supply of AI to cybercriminals has created a stunning variety of potential threats.
Microsoft’s new safety instruments will likely be obtainable for preview on November 3, the corporate has introduced, coming into the more and more crowded discipline of AI cybersecurity options. Earlier this 12 months, Anthropic launched Mythos, a safety platform launched to small accomplice organizations by a program referred to as Glasswing. OpenAI additionally introduced its personal safety answer in Might by a program referred to as Dawn.
In case you purchase by hyperlinks in our articles, we could earn a small fee. This doesn’t have an effect on editorial independence.
