Microsoft Launches MAI-Cyber-1-Flash, Its First Cybersecurity Model, With Agentic Security Platform Perception
Microsoft has launched MAI-Cyber-1-Flash, its first cybersecurity-specialized model, alongside Perception, an agentic platform for automated threat detection and remediation. The company claims the model outperforms rivals from Anthropic, Google and OpenAI on the Cyber Gym benchmark, though no independent scores have been published.
Microsoft on Monday launched MAI-Cyber-1-Flash, its first cybersecurity-specialized AI model, alongside a new agentic security platform called Perception, escalating its competition with Anthropic, Google, and OpenAI in AI-driven cyberdefense.
The announcement came at a small event in San Francisco. According to Microsoft, MAI-Cyber-1-Flash is built "to find challenging vulnerabilities in complex codebases." The model powers MDASH, Microsoft's existing harness for software vulnerability identification and remediation.
What Microsoft claims
Mustafa Suleyman, CEO of Microsoft AI and co-founder of DeepMind, said the company paired MAI-Cyber-1-Flash with GPT-5.4 inside the MDASH harness and that the combination "beats out Gemini, GPT 5.5 Cyber, GPT 5.6 Sol, and Mythos 5 on Cyber Gym" — which Suleyman called the industry's "primary" and "golden" cybersecurity benchmark. Microsoft did not publish specific Cyber Gym scores, pricing, or context window details for the model in the announcement. These performance claims have not been independently verified.
"We're shipping this into production immediately," Suleyman said.
Perception: red, blue, and green teams
The new platform, called Perception, deploys teams of AI agents to automate security workflows across three functions:
- Red teams simulate potential attacks, modeling likely threat actors and the vulnerabilities they might exploit.
- Blue teams detect and triage existing bugs.
- Green teams take corrective action on identified vulnerabilities, including code fixes.
Perception can integrate directly with MDASH. Hayete Gallot, Microsoft's vice president for security, framed the platform as a way for enterprise defenders to "defend against AI with AI at the scale and speed that the attackers have."
Dave Weston, lead engineer for Perception, said the platform compresses work that previously required "hours and hours of manual work from multiple specialized folks" — including appsec hunters and remediation engineers — into minutes, covering discovery, prioritization, detection, posture fixing, and code fixes in one pipeline.
Competitive landscape
Microsoft's launch follows moves by rivals earlier in 2026. Anthropic released a security platform called Mythos to a limited set of partner organizations through a program named Glasswing. OpenAI launched its own cybersecurity solution in May through a program called Day Break. Suleyman's comments explicitly positioned MAI-Cyber-1-Flash against Mythos 5 and OpenAI's GPT 5.5 Cyber and GPT 5.6 Sol models on the Cyber Gym benchmark, though none of these comparative results have been independently confirmed.
Microsoft said both MAI-Cyber-1-Flash and Perception will be available in preview starting November 3, 2026. Pricing has not yet been disclosed.
What this means
Microsoft's entry intensifies a race among the largest AI labs to control the cybersecurity stack — both defensive and, implicitly, offensive tooling that could also inform attackers. The company's claims of benchmark superiority on Cyber Gym are notable but unverified; independent security researchers will need to test MAI-Cyber-1-Flash against real-world vulnerability corpora before the performance claims can be trusted. The bigger signal is structural: Microsoft, Anthropic, and OpenAI are now all shipping agentic security products within months of each other, suggesting security automation has become a priority battleground rather than a niche application. Enterprises evaluating these tools should treat vendor benchmark claims — including Microsoft's — as marketing until third-party validation exists.
Related Articles
Nanbeige4.3-33B released: 33B open model with 512K context targets agentic coding and cowork tasks
Nanbeige has released Nanbeige4.3-33B, a 33B-parameter BF16 model with a 524,288-token (512K) context length. It is the flagship teacher model of the Nanbeige4.3 series and will anchor distillation into smaller models such as Nanbeige4.3-3B. Nanbeige claims state-of-the-art agentic performance among open models in the ~30B class.
KRAFTON releases Raon-1 mini preview, a 28.8B-parameter MoE model with 4.3B active and 262,144-token context
KRAFTON AI has released Raon-1 mini preview, a Mixture-of-Experts language model trained from scratch with 28.8B total and 4.3B active parameters. It is Apache 2.0 licensed, supports 262,144 tokens of context, and ships with a speculative decoding draft model that KRAFTON says reaches up to 2.18× higher output throughput on a single NVIDIA B300.
NVIDIA releases Nemotron 3.5 Bolt E4B 4GB, an open small model targeting 4GB of VRAM
NVIDIA released Nemotron 3.5 Bolt E4B 4GB, an open small language model post-trained from Gemma 4 E4B. It targets local voice assistants, game characters and robots, with a footprint of about 4GB of VRAM on RTX and Jetson hardware.
iFLYTEK Healthcare Upgrades Spark Medical Model to V3.7, Claims 99.6 on MedBench 5.0
iFLYTEK Healthcare announced on October 22 a full upgrade of its Spark medical large model to V3.7, and reports a score of 99.6 on MedBench 5.0. The company says the model now combines speech, medical imaging and language understanding for clinical agent use. The benchmark and usage figures are company-reported.
Comments
Loading...