Mattermost, Cohere bring sovereign AI agents to air-gapped defense ops

The partnership is framed around a specific constraint: mission-critical organizations should not have to send their most sensitive data to a public cloud just to get cutting-edge AI performance. Cohere North was built for private deployment, and pairing it with Mattermost's sovereign operations platform is meant to let even the most tightly regulated customers use top-tier AI without leaving their own infrastructure. Mattermost's Jason Blais argues operators need more than single-player AI — they need a multi-user, multi-agent environment tied to real-time information, with not a single byte of data leaving their control.

On the operational side, the two companies describe controls aimed at exactly that: zero-trust and attribute-based access control (ABAC) policies governing messages, files, agents and workflows, extended to meet national and jurisdictional requirements. The stated use cases range from defending against increasingly sophisticated agentic cyberattacks — connecting SIEM and SOAR platforms to automated incident response without exposing sensitive data to third-party AI — to DevSecOps workflows where engineers and agents work side by side while source code stays in-house. Mattermost also positions the offering as complementary to existing Microsoft Teams and Azure AI Foundry investments, with Mattermost embedded in Teams through the Microsoft Marketplace, including GCC High, so sensitive operations can run fully on-premises within the same Teams experience.

Cohere's Michael McMillan frames the same point from the model side: organizations should not have to ship their most sensitive data to the public cloud solely to obtain state-of-the-art AI. Cohere, founded in 2019, says it has raised about $1.6 billion from investors including Nvidia, AMD Ventures, Salesforce Ventures, Oracle and Cisco, along with AI pioneers such as Geoffrey Hinton and Fei-Fei Li.