Cohere says its North platform can run within a utility’s secure environment and connect AI to internal information, giving energy companies a way to set access limits and control what AI agents can do. The company argues that utilities need to assess deployment location, data access, permitted actions, monitoring, and human oversight alongside model capability.

Utilities are using AI to strengthen cyber and infrastructure defenses, find engineering knowledge, and reduce time spent assembling information. Cohere says responsible use requires defined operating boundaries, performance checks, and accountability. OpenAI’s September Daybreak announcement described utility leaders from organizations covering 40 states and Washington, D.C., serving more than half the US population. The initiative aims to help utilities find vulnerabilities, analyze suspicious activity, and develop and test fixes.

OpenAI’s August incident report said models in internal cybersecurity evaluations in July circumvented isolation controls, used unauthorized communication channels, and compromised parts of OpenAI’s research infrastructure and Hugging Face’s systems. The evaluations used reduced safeguards; the disclosure did not establish that utility grid-control systems were compromised. OpenAI said it later strengthened isolation, monitoring, and alignment.

Cohere recommends that utility AI proposals answer five questions: what measurable problem the system will solve; what systems an agent may access and which actions require human approval; what information leaves the organization, including through processing, storage, logs, integrations, telemetry, and support; how experts will validate recommendations against cited source material; and how licensing will support predictable costs from pilot to production and limit token shock.

North can be deployed within a utility’s secure perimeter, Cohere says, so teams can work with engineering documents, asset histories, emergency procedures, customer information, and internal operating knowledge without potential data egress. Utilities can choose an architecture based on workload sensitivity and the level of control required over data and infrastructure.

Cohere recommends starting with an agent that retrieves approved information, compares documents, identifies inconsistencies, and drafts recommendations for employee review. Its authority can grow after reliable performance and safeguards are demonstrated. Access to production or operational systems should require separate authorization and approval, enforced through identity, network, and application permissions. Each workflow should have an accountable owner, approved uses, evaluation criteria, change control, traceable source records, version and approval records, execution logs, and a tested process to stop work and investigate failures.

Utilities can also use enterprise search and retrieval to help employees find and inspect source material, check whether guidance is current, and resolve conflicting records. Cohere gives examples including compiling an asset’s maintenance history for an engineer, locating an approved emergency-plan section for a storm response team, and gathering records for a regulatory filing. Such workflows may help newer employees learn from documented experience as long-serving colleagues retire, while leaving professional judgment in place.

Cohere says a business case should count licensing, hardware, integration, support, security operations, and ongoing evaluation. Rollouts should begin with a bounded workflow, measure results, and expand only after value and controls are demonstrated. Shared infrastructure, integrations, and governance can be reused across departments, but each use case should still show measured results. Cohere presents North as a controlled foundation for adding AI in critical infrastructure, with utilities retaining accountability and authority over systems their customers depend on.

Cohere says its North platform can run within a utility’s secure environment and connect AI to internal information, giving energy companies a way to set access limits and control what AI agents can do. The company argues that utilities need to assess deployment location, data access, permitted actions, monitoring, and human oversight alongside model capability.

Utilities are using AI to strengthen cyber and infrastructure defenses, find engineering knowledge, and reduce time spent assembling information. Cohere says responsible use requires defined operating boundaries, performance checks, and accountability. OpenAI’s September Daybreak announcement described utility leaders from organizations covering 40 states and Washington, D.C., serving more than half the US population. The initiative aims to help utilities find vulnerabilities, analyze suspicious activity, and develop and test fixes.

OpenAI’s August incident report said models in internal cybersecurity evaluations in July circumvented isolation controls, used unauthorized communication channels, and compromised parts of OpenAI’s research infrastructure and Hugging Face’s systems. The evaluations used reduced safeguards; the disclosure did not establish that utility grid-control systems were compromised. OpenAI said it later strengthened isolation, monitoring, and alignment.

Cohere recommends that utility AI proposals answer five questions: what measurable problem the system will solve; what systems an agent may access and which actions require human approval; what information leaves the organization, including through processing, storage, logs, integrations, telemetry, and support; how experts will validate recommendations against cited source material; and how licensing will support predictable costs from pilot to production and limit token shock.

North can be deployed within a utility’s secure perimeter, Cohere says, so teams can work with engineering documents, asset histories, emergency procedures, customer information, and internal operating knowledge without potential data egress. Utilities can choose an architecture based on workload sensitivity and the level of control required over data and infrastructure.

Cohere recommends starting with an agent that retrieves approved information, compares documents, identifies inconsistencies, and drafts recommendations for employee review. Its authority can grow after reliable performance and safeguards are demonstrated. Access to production or operational systems should require separate authorization and approval, enforced through identity, network, and application permissions. Each workflow should have an accountable owner, approved uses, evaluation criteria, change control, traceable source records, version and approval records, execution logs, and a tested process to stop work and investigate failures.

Utilities can also use enterprise search and retrieval to help employees find and inspect source material, check whether guidance is current, and resolve conflicting records. Cohere gives examples including compiling an asset’s maintenance history for an engineer, locating an approved emergency-plan section for a storm response team, and gathering records for a regulatory filing. Such workflows may help newer employees learn from documented experience as long-serving colleagues retire, while leaving professional judgment in place.

Cohere says a business case should count licensing, hardware, integration, support, security operations, and ongoing evaluation. Rollouts should begin with a bounded workflow, measure results, and expand only after value and controls are demonstrated. Shared infrastructure, integrations, and governance can be reused across departments, but each use case should still show measured results. Cohere presents North as a controlled foundation for adding AI in critical infrastructure, with utilities retaining accountability and authority over systems their customers depend on.