TechnologyBreaking
Archived — This article has been archived. The information may be outdated.
Tech giant unveils first cybersecurity AI model
Tech in Asia··28 Jul
Microsoft unveiled its first cybersecurity specialised AI model, MAI-Cyber-1-Flash, along with a new agentic security platform called Project Perception at an event in San Francisco on Monday. The model powers MDASH, an existing multi-model harness that coordinates more than 100 specialised AI agents to find and fix software vulnerabilities. Project Perception enters public preview on August 3.
Prism
What It Means For You
- Businesses and individuals whose data sits on Microsoft powered systems could see faster detection and patching of security flaws.
- IT professionals evaluating security tools gain a new benchmark to compare against offerings from Anthropic, Google and OpenAI.
- Lower cost AI powered defence tools could eventually make stronger cybersecurity more affordable for smaller organisations too.
What's Happening
- Microsoft launched MAI-Cyber-1-Flash, its first cybersecurity specialised AI model, alongside an agentic platform called Project Perception.
- The model powers MDASH, a harness that coordinates more than 100 specialised AI agents to find and fix software vulnerabilities.
- Project Perception uses red, blue and green agent teams to simulate attacks, triage risks and apply fixes, entering preview on August 3.
A Benchmark Race Between Rival AI Labs
- Microsoft said its model scored about 96 percent on the CyberGym benchmark, 12 points above Anthropic's Mythos platform.
- The configuration cuts the cost of running the MDASH harness by roughly half by routing only the hardest tasks to a larger model.
- High impact actions taken by Perception's agents still require human sign off before deployment, Microsoft said.
all-newseditors-picks
