Show the Audit Trail: Utility AI Meets Regulators, Cost Recovery and DISCOM Data

Utility AI is moving from pilot slides to filings, audits and rate cases. Recent reporting shows the same pattern on both sides of the Atlantic and in India: narrow tools that forecast, prioritise and plan are spreading, while the questions regulators now ask are about auditability, data quality and who pays.

Narrow tools are working; autonomy is not here

A Utility Dive report from 22 October 2026 describes most utility AI as narrow systems that support human decisions, such as load forecasting and vegetation management. Utilities are far from letting large language models act on physical infrastructure. National Grid Partners found that nearly 80% of utility leaders in the US, UK and Europe had fully deployed at least one AI application for large-load customer planning, yet adoption is strongest in function-specific tools rather than company-wide platforms.

One concrete example in the same report: Portland General Electric, working with GridCARE and hourly demand modelling, identified more than 80 MW of data centre connection capacity years earlier than expected. The pattern is useful: AI delivered value by sharpening a planning question that already existed.

A separate Utility Dive opinion piece from an AWS energy executive, published 15 October 2026, argues that AI can compress interconnection studies from months to hours. It cites Duke Energy doubling the number of projects evaluated in a year and PG&E merging data from more than 80 sources across 20 systems to prioritise over 17,000 grid upgrades. Because the author is a vendor executive, these figures are best treated as claims to verify, though the stated principle stands: tools recommend, and human engineers decide.

The governance gap is now a regulatory topic

The same Utility Dive report notes that Guardrail Technologies reviewed federal filings and found about 70% of energy and utility companies identified AI as a cybersecurity risk, while only 23% of utilities documented a process for managing it. FERC's early October 2026 cybersecurity audit flagged growing AI use and warned of access control risks. New York regulators directed utilities to file inventories of AI use, and Connecticut's Attorney General and Consumer Counsel petitioned regulators on 13 October for a similar review.

Guardrail's chief executive is quoted saying most utilities could not reconstruct what an AI system read, what it did and why. Render Networks' chief executive put the data point bluntly: automating against bad data only produces the wrong answer faster.

Cost recovery decides what gets built

E&E News reported in May 2026 that regulators' repeated rejection of cost recovery has slowed adoption of AI-based grid tools. PG&E is making the case for roughly $45 million for a wildfire monitoring programme involving 5.5 million meters, 1,600 AI-enabled weather stations and 600 infrared cameras, and state regulators are resisting. The article also notes past denials in North Carolina, Hawaii and Illinois for software and cloud costs. Consumer advocates worry about gold-plating, and one data executive observed that regulators have not asked for quantitative, risk-based cost-benefit analysis.

The National Grid Partners survey in the October report points the same way: 87% of respondents said outdated regulatory and rate case frameworks hinder returns on AI investment.

India: policy intent, analytics scope

India's framing is similar but earlier in the cycle. At the first national conference on AI and ML in power distribution, reported by News On AIR on 7 December 2025, Power Minister Manohar Lal named smart meter analytics, GenAI-based decision support and demand forecasting among the applications that can reduce DISCOM losses and power purchase costs. The conference also saw the launch of STELLAR, a software tool for DISCOM resource adequacy plans.

At state level, an MoU between ISGF and GUVNL, covered by Energetica India after its 1 May 2026 exchange, targets AI/ML-based analytics for Gujarat's DISCOMs, including demand forecasting and reduction of technical and commercial losses, with pilots and scalable deployment. The article reports no numeric targets, budgets or deadlines, which is itself informative: the measurable outcome still has to be defined.

What it means for utilities

• Data quality comes before model quality. Forecasting and loss analytics inherit every error in meter, GIS and consumer-index data.
• An AI inventory is becoming a basic requirement. Regulators in New York and Connecticut are asking what is deployed. DISCOMs and their regulators can expect similar questions.
• Audit trails are part of the product. If a model's inputs, outputs and actions cannot be reconstructed, scaling it into protection, dispatch or field operations is hard to defend.
• Business cases need regulatory language. Programmes that quantify avoided outages, avoided crew dispatches or deferred capex are better placed to survive tariff scrutiny than general modernisation claims.
• MoUs need outcome metrics. Loss reduction and forecast accuracy targets, with baselines, turn a pilot into an accountable programme.

Key takeaways

• Utility AI today is mostly narrow decision support, not autonomous control.
• About 70% of energy and utility firms flag AI as a cyber risk, but few document how it is managed.
• Cost recovery, not technology, has historically limited adoption of grid analytics.
• Interconnection and large-load planning are the most visible near-term use cases, though vendor claims need independent verification.
• India's policy and state MoUs set direction; published targets and baselines are the missing layer.

Sources

• Utilities turn to AI to manage rising power demand — from AI (Utility Dive)
• Utilities can use AI to solve grid backlogs (Utility Dive)
• AI could make power grid more efficient — if utilities can persuade regulators (E&E News)
• Union Minister Manohar Lal highlights transformative role of AI/ML in power distribution (News On AIR)
• ISGF signs MoU with GUVNL for AI-driven DISCOM transformation in Gujarat (Energetica India)