1. Introduction

Modern electricity grids increasingly rely on automated systems such as SCADA, Energy Management Systems (EMS), Automatic Generation Control (AGC), automatic protection systems, demand-response platforms, automated voltage control, and AI-based forecasting and dispatch tools. These systems can process enormous quantities of data and react much faster than human operators.

However, automation does not eliminate the legal responsibility of the grid operator. The central regulatory principle is that automation should operate within a framework of human accountability, supervision, intervention and override, particularly where an automated decision can materially affect grid stability, electricity supply, market integrity or public safety.

This principle is becoming particularly important as AI moves from advisory functions toward increasingly autonomous grid operations. A 2025–26 UK independent review on AI deployment in electricity networks specifically identifies the need for governance as electricity systems move toward increasing levels of autonomy. (GOV.UK) NERC materials likewise identify human oversight and override capabilities, validation, failover systems and cybersecurity as important considerations for AI-enabled grid operations. (NERC)

2. Meaning of Human Oversight

Human oversight means that a qualified person or institution retains meaningful ability to:

observe the automated system;

understand or evaluate its operational output;

identify abnormal or unsafe decisions;

intervene before or during execution where appropriate;

override or deactivate the automated function;

initiate emergency procedures;

investigate automated decisions afterwards; and

remain accountable for compliance with applicable electricity and reliability rules.

Human oversight therefore differs from merely having a person physically present in a control room.

A system in which an operator is technically present but cannot understand, interrupt or override an automated command would provide substantially weaker oversight.

3. Why Human Oversight Is Legally Important

Electricity networks possess characteristics that distinguish them from many ordinary automated systems.

A disturbance in one part of an interconnected grid can propagate rapidly into other areas. Historical experience, including the 2003 North American blackout, demonstrated the systemic consequences of failures in monitoring and grid control. Congressional testimony concerning grid operations has emphasised that interconnected systems can allow disturbances to cascade across large portions of the network. (GovInfo)

Consequently, automated grid-control systems should generally be designed around five legal objectives:

Safety → Reliability → Accountability → Recoverability → Auditability

4. Sources of Human Oversight Obligations

Human oversight obligations can arise from several different legal sources.

A. Electricity legislation

Electricity statutes generally impose responsibilities on transmission operators, system operators, distribution licensees and load-dispatch authorities.

B. Grid codes

Grid codes establish technical requirements governing frequency, voltage, system security, dispatch, protection and operational coordination.

C. Reliability standards

In the United States, NERC reliability standards provide a major regulatory framework for bulk-power-system reliability, with FERC exercising federal regulatory authority over the framework.

D. Licence conditions

A licence may require an operator to maintain competent personnel, appropriate control systems and emergency arrangements.

E. Administrative law

Where automated decisions affect regulated entities or market participants, requirements concerning reasoned decision-making, procedural fairness and regulatory accountability can become relevant.

F. Tort, negligence and public-law liability

If inadequate supervision of automation contributes to physical or economic harm, ordinary principles of negligence, statutory duty and public-law accountability may become relevant depending on jurisdiction.

5. Indian Legal Framework

India provides a particularly useful example because the Electricity Act, 2003 assigns important operational responsibilities to Load Despatch Centres.

Section 31 establishes the State Load Despatch Centre (SLDC), while Section 32 sets out its functions. These include supervision and control of the intra-State transmission system and ensuring integrated operation of the power system.

Section 33 gives the SLDC authority to issue directions to generating companies, licensees and other participants for maintaining system operation.

A 2026 Chhattisgarh High Court decision discussing the statutory framework expressly described the SLDC's responsibilities as including scheduling and dispatch, monitoring grid operations, maintaining electricity quality and carrying out real-time grid control for secure and economic operation. (Indian Kanoon)

This is significant for automation: delegating a technical operation to an automated system does not necessarily transfer the statutory responsibility of the SLDC or system operator to the machine.

The automation becomes a means through which the statutory duty is performed.

6. Duty to Monitor Automated Decisions

The first major obligation is continuous or appropriately risk-based monitoring.

An operator should be able to determine:

what the automated system is doing;

what data it is using;

what assumptions it is making;

whether its outputs fall within approved operating parameters;

whether abnormal conditions have arisen; and

whether the automated action is producing the expected physical result.

This is particularly important for AI systems because an AI model may produce an operational recommendation without providing the same type of deterministic reasoning traditionally associated with conventional protection or control systems.

Recent research concerning generative AI for power-sector applications therefore describes a human-in-the-loop architecture in which operators review model outputs, evaluate scenarios and make the final operational decision, with actions logged and subsequently audited. (DOI)

7. Duty to Maintain Human Override

A particularly important obligation is the ability to override automation.

An automated grid-control system should have clearly defined mechanisms for:

manual intervention;

emergency shutdown;

disabling automated commands;

reverting to a safe operating mode;

switching to backup control systems; and

transferring control between authorised operators.

The appropriate level of override depends upon the risk of the automated function.

For example:

This illustrates an important distinction: human oversight does not necessarily mean that a human must approve every millisecond-level protection action.

Where physical events occur faster than human reaction time, autonomous protection may be legally and technically necessary. Human responsibility then shifts toward system design, authorisation, testing, supervision and post-event review.

8. Duty to Establish Intervention Thresholds

Operators should establish beforehand when human intervention becomes mandatory.

Examples include:

frequency deviation beyond prescribed limits;

unexpected voltage instability;

loss of communications;

conflicting sensor readings;

model-confidence deterioration;

unexpected generator behaviour;

cyberattack indicators;

abnormal automated switching;

failure of redundancy;

deviation between predicted and actual system conditions.

The advantage of predetermined thresholds is that the operator does not have to decide during a crisis whether intervention is required.

9. Duty of Competent Supervision

Human oversight is meaningful only if the responsible personnel possess adequate competence.

This creates obligations relating to:

operator training;

AI-system training;

emergency exercises;

understanding of automated-control limitations;

cybersecurity awareness;

simulation exercises;

periodic competency assessments; and

procedures for unusual system conditions.

The legal concept is therefore not simply human-in-the-loop, but competent human-in-the-loop.

10. Duty to Validate Before Deployment

A grid operator should not introduce autonomous control merely because the technology appears technically capable.

Validation should consider:

Technical validation

Whether the system performs correctly under ordinary conditions.

Stress testing

Whether it remains reliable during extreme events.

Failure testing

What happens if sensors, communications, software or models fail.

Adversarial testing

Whether manipulated or misleading data can cause unsafe decisions.

Human-factors testing

Whether operators can actually understand and override the system under time pressure.

Cybersecurity testing

Whether unauthorised parties can manipulate automated control.

NERC-related materials concerning AI adoption specifically identify testing and validation, cybersecurity integration, failover systems and human oversight/override as important implementation considerations. (NERC)

11. Duty to Maintain Fail-Safe and Failover Mechanisms

Human oversight cannot depend upon the assumption that the primary automated system will always function.

A legally defensible architecture should therefore contemplate:

AI/automation failure → alarm → operator notification → safe state → backup control → manual intervention

Depending on the system, this could include:

redundant control servers;

independent protection systems;

backup communications;

manual switching;

emergency operating procedures;

independent measurements;

black-start procedures; and

geographically separated control facilities.

12. Accountability Cannot Simply Be Transferred to AI

A central legal question is:

Who is responsible when an automated grid-control system makes a harmful decision?

The machine itself is generally not the regulated electricity entity.

Responsibility may instead be examined at several levels:

System owner → operator/licensee → control-room personnel → software developer/vendor → maintenance provider → regulator

The relevant question becomes whether the responsible organisation:

selected an appropriate system;

properly tested it;

established appropriate operating limits;

adequately supervised it;

maintained an override mechanism;

responded appropriately to warnings; and

complied with applicable reliability requirements.

13. Case Law

A. Austrian Power Grid AG and Others v ACER, T-606/20

The General Court of the European Union decided this case on 15 February 2023 concerning the European electricity-balancing framework and automatic frequency-restoration reserves (aFRR). The litigation concerned ACER's authority and the legal framework governing implementation of European balancing platforms. (Infocuria)

Although this was not an AI-liability case, it is important for automated grid control because it demonstrates that technically sophisticated and highly automated electricity-balancing arrangements remain subject to legally defined institutional authority and procedural requirements.

Principle:
Automation does not remove the underlying regulatory allocation of authority between transmission system operators and regulatory institutions.

B. Polskie Sieci Elektroenergetyczne SA and Others v ACER, Joined Cases C-281/23 P and C-282/23 P

The Court of Justice's 23 October 2025 judgment concerned the European platforms for balancing energy, including platforms involving automatic activation (aFRR) and manual activation (mFRR). (Infocuria)

The case demonstrates the continuing importance of legally defined governance even where electricity balancing is increasingly automated.

Relevance to human oversight: automated balancing operates within a regulatory architecture; it does not constitute an independent legal authority capable of replacing the institutions responsible for grid governance.

C. Jindal Steel and Power Ltd v Chhattisgarh State Electricity ... (2026)

This 2026 Indian decision is particularly useful for the Indian context because the court discussed the statutory responsibilities of the SLDC under the Electricity Act, 2003.

The judgment records the statutory framework under which the SLDC is responsible for integrated system operation, scheduling and dispatch, monitoring grid operations and real-time grid control. (Indian Kanoon)

Legal significance:
Where an SLDC uses automated systems for operational control, those systems should be understood as instruments supporting the SLDC's statutory responsibilities rather than replacements for the legal institution itself.

D. Citadel FNGE Ltd v FERC, 94 F.4th 362 (D.C. Cir. 2023)

This case concerned electricity-market pricing and congestion management within PJM. The D.C. Circuit's decision illustrates the close regulatory scrutiny applied to PJM's management of electricity markets and congestion. (FindLaw)

Although the case does not establish a general AI human-oversight rule, it is relevant because automated grid and market-control systems operate within regulated market structures. Algorithmic or automated decision-making therefore remains subject to applicable FERC and market rules.

14. Emerging FERC/NERC Position

The regulatory direction in the United States increasingly recognises that AI and automation must be integrated without compromising reliability.

In December 2025, FERC directed PJM to establish transparent rules addressing the connection of AI-driven data centres and other large loads, explicitly linking technological development with grid reliability and consumer protection. (Federal Energy Regulatory Commission)

Separately, NERC-related materials identify human oversight and override capabilities as considerations for AI-based grid operations. (NERC)

This suggests a regulatory model in which increasing automation is compatible with grid operation only when accompanied by appropriate governance, reliability controls and accountability.

15. Human Oversight and AI-Based Grid Control

AI introduces additional issues beyond traditional automation.

An AI model may:

generate an incorrect forecast;

misinterpret unusual operating conditions;

respond differently to previously unseen circumstances;

rely on corrupted or incomplete data;

produce an apparently plausible but unsafe recommendation;

become unreliable after changes to its underlying model; or

interact unpredictably with other automated systems.

Accordingly, AI-based grid control should preferably use a graduated autonomy model:

Level 1 — Advisory

AI provides information or recommendations.

Human: makes the operational decision.

Level 2 — Supervised automation

AI performs routine operations within predetermined boundaries.

Human: continuously supervises and can intervene.

Level 3 — Conditional autonomy

AI operates independently during predefined circumstances.

Human: establishes conditions and retains emergency override.

Level 4 — High autonomy

AI performs complex operational actions with limited real-time intervention.

Human: remains responsible for system governance, safety architecture, testing, monitoring and intervention capability.

The greater the potential physical consequences, the stronger the justification for robust human supervision and intervention mechanisms.

16. Record-Keeping and Auditability

Another important obligation is maintaining records of automated decisions.

A suitable audit trail should identify:

time of automated decision;

system version;

input data;

operator on duty;

automated recommendation;

action executed;

warnings generated;

human intervention;

override attempts;

resulting grid condition; and

subsequent corrective measures.

This is important because after a blackout or other incident, investigators must be able to reconstruct what the automated system did and what humans knew at the relevant time.

17. Liability for Failure of Human Oversight

Potential liability can arise from several failures.

Failure 1: No monitoring

The operator allowed an automated system to operate without meaningful supervision.

Failure 2: Failure to intervene

The operator received a warning but failed to take appropriate action.

Failure 3: Inadequate design

The system was deployed without a functional override.

Failure 4: Inadequate training

Personnel could not understand or safely operate the automated system.

Failure 5: Inadequate validation

The system was deployed without sufficient testing against foreseeable grid conditions.

Failure 6: Excessive automation

Critical decisions were delegated to automation without appropriate risk controls.

Failure 7: Poor cybersecurity

The operator failed to protect the automated control system against unauthorised manipulation.

Thus, liability may arise not because automation itself was unlawful, but because the organisation failed to establish an adequate governance structure around the automation.

18. Core Legal Principle

The emerging principle can be expressed as follows:

Automation may perform the operational function, but responsibility for lawful and reliable operation remains attached to the legally responsible human institution and organisation.

This produces a distinction between:

Machine autonomy
and
legal accountability.

They are not necessarily the same thing.

19. Proposed Legal Framework

A comprehensive regulatory framework for automated grid control should require:

Risk classification of automated functions.

Mandatory human oversight for high-consequence functions.

Clearly defined intervention thresholds.

Human override mechanisms.

Independent fail-safe mechanisms.

Pre-deployment validation.

Periodic revalidation after software/model changes.

Operator training and certification.

Comprehensive event logging.

Cybersecurity controls.

Incident reporting.

Post-event investigation.

Vendor accountability provisions.

Regulatory audit rights.

Clear allocation of responsibility between system operator and technology provider.

20. Conclusion

Human oversight obligations for automated grid control arise from the fundamental character of electricity as a critical infrastructure system. Automated control can improve speed, precision and reliability, but it does not eliminate the legal duties of system operators, transmission organisations, distribution licensees or load-dispatch authorities.

Indian law is particularly relevant because the Electricity Act, 2003 places operational responsibilities directly upon Load Despatch Centres. Recent Indian case law continues to emphasise the statutory role of these institutions in maintaining secure and coordinated grid operation. (Indian Kanoon)

European electricity-balancing litigation similarly demonstrates that increasingly automated balancing arrangements remain embedded within legally defined institutional and procedural structures. (Infocuria)

The emerging legal model is therefore not “human versus automation.” It is automation under accountable human governance: machines may execute increasingly sophisticated grid functions, but humans and regulated institutions must retain appropriate powers of supervision, intervention, override, investigation and responsibility.