What happened
GovTech reports that Connecticut’s attorney general and consumer counsel asked the Public Utilities Regulatory Authority to investigate utilities’ AI systems, their purposes, human oversight, access rights and security controls. The request follows a Federal Energy Regulatory Commission audit warning that AI tools already present in some protected utility environments could complicate detection and enable unauthorized network movement.
GovTech reports that Connecticut Attorney General William Tong and Consumer Counsel Claire Coleman petitioned PURA to determine where electric, gas and water utilities use AI, why they use it, what systems it can access and whether existing safeguards are adequate. The proposed review would cover storm response, customer systems, asset management, energy-load and water-demand forecasting, procurement, grid and power-plant operations, maintenance, third-party vendors and advanced metering infrastructure.
According to GovTech, Eversource and Avangrid have publicly acknowledged using AI, but the petition says the extent and scope of those applications remain unclear. The petition asks utilities to identify their systems, explain each system’s objective, describe human oversight and provide relevant policies and procedures. Eversource said it supports a transparent evaluation but did not provide details about its current applications.
GovTech also reports that FERC’s 2026 cybersecurity audit described increasing use of chatbots, personal assistants and automated analytical systems in protected environments associated with bulk-power reliability requirements. FERC warned that insufficiently authorized, verified or monitored tools could mimic normal user behavior, complicate incident detection and create unexpected routes toward protected systems. FERC did not identify the audited entities, and the report as supplied does not establish that any Connecticut utility experienced a related incident.
The report says the Connecticut petition cites concerns about AI weather forecasting and acknowledges uncertainty about whether utilities in the Northeast used AI models during a February nor’easter. Researchers quoted by GovTech distinguish between AI that assists an engineer and AI that takes an action affecting the grid, emphasizing that human oversight is meaningful only when people have enough information, time and authority to intervene.
Why it matters
AI is moving into systems that support electricity, gas and water services, but the report says Connecticut officials do not yet know the full scope of that use. The practical risk is not limited to spectacular failures: an AI system with excessive permissions, poor emergency performance or access to sensitive smart-meter data could affect reliability, privacy or consumer protections. GovTech reports no AI-caused breach or operational failure in Connecticut, and the cited concerns have not been independently confirmed here.
GovTech’s reporting connects two oversight questions: whether AI is secure inside utility networks and whether it is reliable when conditions depart from normal operations. A model that performs well during ordinary demand or weather may behave differently during an emergency, when incorrect forecasts or recommendations could carry greater consequences. The article presents these as risks requiring investigation, not as demonstrated failures.
The consumer implications extend beyond grid control. GovTech reports that a customer chatbot could give a confidently incorrect answer about a bill, payment arrangement, shutoff protection or restoration estimate. The petition’s inclusion of smart meters also raises a privacy question: detailed electricity-use patterns may reveal household routines or occupancy, particularly when data is processed by outside vendors.
The report does not independently test any utility AI system, identify a specific vulnerability, establish a breach or quantify customer harm. It also does not establish that all utilities use AI in operationally sensitive functions. Those limits matter because the requested review is intended to discover the systems, permissions and safeguards that are currently unclear.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').Why can ethical evaluation not be reduced to one model score?
What to watch next
The key next step is whether PURA opens the requested review and requires utilities to disclose their AI inventory, safeguards, testing and accountability arrangements. Watch for evidence about emergency testing, vendor access, human intervention authority, customer-facing chatbot controls and the handling of advanced-metering data.
PURA’s response will indicate whether Connecticut utilities must produce a system-by-system inventory or only general descriptions. Useful disclosures would identify each system’s role, data sources, permissions, vendor involvement, testing methods, fallback procedures and accountable decision-makers without exposing sensitive technical details.
GovTech reports that FERC auditors found utility compliance programs were slower to adapt than vendors adding AI capabilities to products and services. Future reporting should clarify how utilities review vendor updates, restrict tool permissions, monitor activity and determine whether newly added capabilities alter the security or reliability risk.
For customer-facing uses, watch for testing on high-stakes questions and a clear path to a human representative. For operational uses, watch for evidence of testing under extreme weather, unusual demand, degraded data and other emergency conditions. The supplied report provides no access or pricing information because it concerns utility oversight rather than a public AI product.