Updated
Updated · CNBC · Jul 27
Microsoft Unveils 50%-Cheaper MAI-Cyber-1-Flash as It Revives $20 Billion Security Business
Updated
Updated · CNBC · Jul 27

Microsoft Unveils 50%-Cheaper MAI-Cyber-1-Flash as It Revives $20 Billion Security Business

3 articles · Updated · CNBC · Jul 27

Summary

  • MAI-Cyber-1-Flash is Microsoft's first cybersecurity-specific generative model and will enter Project Perception's public preview on Aug. 3 to find and help fix software vulnerabilities.
  • Paired with OpenAI's GPT-5.4, the model beat Anthropic's Mythos 5, Google's 3.5 Flash Cyber and OpenAI's GPT-5.5 Cyber on the CyberGym benchmark, which Microsoft said delivers world-leading performance at half the cost.
  • Hayete Gallot's February return to lead security marks Microsoft's first major push to revive the unit since Charlie Bell stepped back, with Perception designed to connect to non-Microsoft tools and implement approved code changes.
  • The launch also supports Satya Nadella's push for more in-house AI as Microsoft shares are down 19% in 2026 and investors worry its heavy OpenAI exposure could become a risk amid rising open-source competition.
  • Microsoft last disclosed cybersecurity revenue in 2023 at more than $20 billion, and executives argue AI defenders are needed as attackers increasingly use generative models to exploit newly disclosed flaws.

Insights

If AI can unexpectedly hack systems during tests, what stops autonomous AI defenders from going rogue inside your network?
As AI evolves from reading data to executing actions, are we deploying the ultimate cyber shield or the perfect weapon?
When malicious instructions hide in trusted metadata, how can businesses trust AI agents to hunt threats without silently stealing data?