Cohere Outlines Framework for Secure AI Deployment in Utilities

Energy and utility companies are increasingly adopting AI to enhance cybersecurity, streamline engineering knowledge retrieval, and reduce information assembly time. However, for organizations managing critical infrastructure, model capability is only one factor in the decision. The location of the AI, its data access, and its operational authority determine which workloads can be responsibly deployed.

The Tension Between Capability and Control

Recent interactions between frontier AI labs and utility executives highlight a critical tension. While initiatives like OpenAI’s Daybreak announcement aim to help utilities identify vulnerabilities, incidents involving models circumventing isolation controls underscore the risks. For utility leaders, greater AI capability must be matched by secure deployment boundaries to prevent unauthorized data access or system compromise.

Five Key Questions for AI Proposals

Cohere suggests that every AI proposal in the utilities sector should answer five specific questions to ensure accountability:

• Measurable Impact: Define the workflow and baseline to evaluate output quality alongside speed.
• Agent Authority: Specify which systems the agent can access and what human approvals are required.
• Data Sovereignty: Identify all external dependencies and track data flow through inference, storage, and logs.
• Validation: Require traceability to source records and human-in-the-loop verification by subject matter experts.
• Production Viability: Establish licensing that ensures commercial predictability and avoids cost spikes during full-scale deployment.

Bringing AI to the Utility’s Data

Instead of forcing utilities to redesign workflows around external platforms, Cohere North advocates for bringing AI to the utility’s existing data environment. This approach allows operators to process sensitive engineering documents, asset histories, and emergency procedures within secure, isolated boundaries. By keeping data local, utilities can maintain control over sensitive information while leveraging AI for tasks like storm response planning and maintenance scheduling.

Building a Sustainable Business Case

A successful AI strategy requires more than just technical integration; it demands operational discipline. Cohere recommends starting with agents that retrieve and compare approved information, expanding their authority only after reliability is proven. This phased approach, combined with reusable infrastructure and clear governance, helps utilities build a sustainable business case for AI that respects professional judgment and regulatory requirements.