Cohere offers to put agents in lockdown mode with strict ACLs

You can now restrict which proprietary data specific employees can access when using AI agents.

Cohere has updated its North agent harness, now called North 2, to enhance enterprise AI agent security and collaboration. Launched in early 2025, the North agent harness provided tools for automating business processes by connecting to existing knowledge bases and SaaS applications. The revamped North 2 introduces "skills," which are documented standard operating procedures for agent tasks, designed to be shareable to avoid redundant work and ensure consistency. It also includes "automations," customizable templates created with a drag-and-drop interface, and "libraries" for shared knowledge and assets to ensure agents operate on consistent datasets. North 2 incorporates memory to maintain context across sessions and offers deployment flexibility, allowing on-premise or cloud installation, including in air-gapped environments. New guardrails aim to control token spend and enforce access controls, with flow control for visibility into departmental token usage and access control lists for libraries and skills to restrict data access.

Cohere's North 2 update addresses enterprise concerns about AI agent security and cost control by introducing robust access controls and token spend management. The emphasis on "skills" and "libraries" suggests a move towards more standardized and reusable AI workflows, akin to self-generating macros that operate across multiple applications. This focus on consistency and security positions Cohere as a strong contender in the increasingly crowded enterprise agentic AI market, particularly for organizations prioritizing data governance and predictable operational costs.

By offering features like access control lists for libraries and skills, Cohere aims to provide enterprises with the granular control needed to deploy AI agents safely within regulated environments. This is particularly relevant as other major players like OpenAI and Anthropic also navigate the challenges of developing safer, more controllable AI agents, while adhering to enterprise demands for sovereign AI deployment options. The competitive pressure in this space is escalating, driving innovation in agent security and management functionalities.