1. Introduction

The increasing digitalisation of electricity systems is transforming the way energy is generated, transmitted, distributed and consumed. Modern grids increasingly use artificial intelligence (AI), machine learning, automated forecasting, digital twins, predictive maintenance, automated demand response, smart meters and algorithmic dispatch systems. These technologies can process enormous volumes of data and make recommendations or operational decisions much faster than human operators.

However, electricity is a critical infrastructure system. An erroneous automated decision can cause equipment damage, service interruption, market distortion, safety risks or unequal impacts on consumers. Consequently, a governance model in which AI operates entirely without human intervention raises important questions concerning accountability, transparency, safety and legal responsibility.

A Hybrid Human–AI Governance Model attempts to combine computational capabilities of AI with human legal and institutional responsibility. Under such a model, AI may monitor, predict, optimise and recommend, while humans retain defined powers of supervision, intervention, review and accountability.

This approach is particularly relevant in India because the Electricity Act, 2003 establishes institutional responsibilities for grid standards, safety and system operation. The Central Electricity Authority (CEA), for example, has statutory functions concerning technical standards, safety requirements and grid standards under the Act. (Central Electricity Authority)

2. Meaning of Hybrid Human–AI Governance

Hybrid Human–AI governance can be defined as:

A regulatory and operational framework in which AI systems perform specified analytical, predictive or operational functions while human institutions retain legally defined oversight, intervention and accountability powers.

The model therefore rejects two extremes:

Pure human governance – all important decisions are made manually; and

Fully autonomous AI governance – AI systems make consequential decisions without meaningful human oversight.

Instead, authority is distributed between humans and machines.

Typical structure

Data → AI system → Prediction/Recommendation → Human review → Authorised decision → Execution → Monitoring → Audit

For example, an AI system may predict that electricity demand will rise sharply during the evening. It can recommend additional generation or demand-response measures. A system operator can examine the recommendation, consider grid-security requirements and authorise the action.

3. Why Hybrid Governance Is Necessary in Energy Systems

A. Complexity of modern electricity networks

Modern electricity systems contain millions of data points from:

smart meters;

renewable-energy plants;

batteries;

electric vehicles;

transmission lines;

substations;

weather systems;

demand-response platforms; and

distributed energy resources.

Human operators cannot independently process all of this information in real time.

AI can therefore provide substantial assistance.

B. Critical infrastructure risk

Unlike many ordinary commercial applications, electricity infrastructure has systemic consequences. A defective AI recommendation affecting a transmission network could potentially propagate through interconnected infrastructure.

The CEA's statutory framework specifically includes technical standards, safety requirements and grid standards. (Central Electricity Authority)

Therefore, AI governance must operate within existing electricity-sector safety and reliability obligations.

C. Accountability

If an AI-controlled system makes an erroneous decision, a basic legal question arises:

Who is responsible?

Possible candidates include:

the utility;

system operator;

AI developer;

equipment manufacturer;

data provider;

regulator; or

human decision-maker who approved the AI recommendation.

Hybrid governance attempts to prevent an accountability gap by assigning responsibility before the system is deployed.

4. Core Components of a Hybrid Human–AI Governance Model

4.1 Human-in-the-loop

Under a human-in-the-loop model, an AI system can analyse information and generate recommendations, but a human must approve important decisions.

Example:

AI predicts transmission congestion → recommends redispatch → system operator reviews → operator authorises redispatch.

This model is appropriate for decisions with substantial safety, economic or public-interest consequences.

4.2 Human-on-the-loop

Under human-on-the-loop governance, AI can execute predetermined actions automatically, while a human continuously supervises the system and can intervene.

For example, an automated protection system may disconnect a component during an emergency while the system operator monitors the event.

This is particularly relevant where decisions must occur faster than humans can respond.

4.3 Human-over-the-loop

At the strategic level, humans determine:

operational objectives;

safety thresholds;

permissible AI functions;

emergency procedures;

risk tolerance;

data-governance requirements; and

accountability arrangements.

AI therefore operates within a framework established by human institutions.

5. Risk-Based Allocation of Human Oversight

Not every AI decision requires the same degree of human intervention.

A useful governance framework can classify decisions into three categories.

The appropriate level of oversight should depend upon potential consequences, not merely whether AI is involved.

6. AI in Electricity Dispatch

AI can assist electricity-system operators with:

generation scheduling;

renewable forecasting;

congestion management;

demand forecasting;

reserve allocation;

battery dispatch;

outage prediction;

voltage optimisation; and

demand response.

The legal difficulty arises when an AI recommendation becomes an operational decision.

For example:

AI recommends shutting down Generator A because of predicted congestion.

A hybrid model requires the system to establish:

why the recommendation was generated;

what data were used;

whether the recommendation satisfies grid rules;

whether an operator must approve it;

who is responsible if it causes harm; and

whether the decision can subsequently be audited.

7. Human Oversight and Indian Electricity Law

India's electricity framework provides an important foundation for hybrid governance.

The Electricity Act, 2003 gives the CEA responsibilities including technical standards for electrical plants and lines, safety requirements and Grid Standards. (Central Electricity Authority)

The CEA's current regulatory framework continues to include categories such as grid standards, grid connectivity, safety and communication regulations. (Central Electricity Authority)

The Regional Power Committees also have responsibilities concerning stable and smooth regional-grid operation, operational planning, protection systems and automatic under-frequency load shedding. (Central Electricity Authority)

These provisions are important for AI governance because AI-based control cannot operate outside the statutory responsibilities of grid institutions.

Legal principle

AI should therefore be treated as an instrument of electricity-system governance, rather than as an independent legal authority.

8. Algorithmic Transparency

One of the principal problems with AI governance is the black-box problem.

A sophisticated machine-learning model may generate a recommendation without providing a simple explanation understandable to the operator.

In critical infrastructure, this creates several risks:

inability to identify errors;

inability to challenge decisions;

difficulty assigning responsibility;

difficulty auditing discriminatory outcomes; and

difficulty reconstructing the causes of an incident.

A hybrid model should therefore require appropriate explainability, particularly for consequential decisions.

This does not necessarily mean that every mathematical detail of a model must be publicly disclosed. Instead, operators and regulators should have sufficient information to determine:

the basis of the recommendation;

relevant inputs;

confidence levels;

applicable constraints;

known limitations; and

circumstances in which human intervention is required.

9. Human Override

A central feature of hybrid governance is the override mechanism.

An operator should be able to suspend or override AI recommendations when:

the system produces an obviously erroneous recommendation;

sensor data appear unreliable;

cybersecurity concerns arise;

unusual grid conditions exist;

emergency conditions occur; or

regulatory requirements conflict with the AI recommendation.

However, the override mechanism must itself be governed.

A system where an operator can arbitrarily override AI decisions without recording reasons can create accountability problems.

Therefore, significant overrides should ordinarily generate an audit trail.

10. AI Confidence Thresholds

AI systems should not necessarily act identically regardless of their confidence level.

For example:

95–100% confidence: automated recommendation/action may be permitted;

70–95%: human confirmation may be required;

below 70%: human investigation may be mandatory.

These numerical thresholds are merely illustrative; actual thresholds must be determined through engineering validation and regulatory risk assessment.

The broader legal principle is:

Greater uncertainty + greater potential harm = stronger human oversight.

11. The Importance of Auditability

Every consequential AI decision should, where technically feasible, generate an audit record containing:

time of decision;

system state;

relevant input data;

AI model/version;

recommendation;

confidence level;

human intervention;

final decision;

outcome; and

subsequent corrective action.

This is particularly important in electricity regulation because regulators may need to reconstruct events after:

blackouts;

equipment failures;

market manipulation;

consumer disputes;

cyber incidents; or

unexpected automated actions.

12. Data Governance

AI governance is fundamentally dependent on data governance.

Energy AI systems can process:

electricity consumption;

household load profiles;

industrial activity;

location information;

weather data;

market transactions; and

equipment information.

Therefore, hybrid governance should address:

Data accuracy

Incorrect data can produce incorrect AI decisions.

Data security

Energy datasets can constitute critical infrastructure information.

Data minimisation

Only necessary information should be processed.

Data integrity

Operators must be able to determine whether data have been altered.

Data provenance

The origin of important datasets should be identifiable.

13. Cybersecurity and Human–AI Governance

AI can also create new cybersecurity risks.

An attacker could manipulate:

training data;

sensor information;

forecasting inputs;

optimisation parameters; or

AI models.

Consequently, cybersecurity must be incorporated into the governance model.

The CEA's current materials include a dedicated Cyber Security Regulations category/update, demonstrating the continuing regulatory significance of cybersecurity within India's electricity framework. (Central Electricity Authority)

A hybrid governance system should therefore include:

authentication;

access control;

model integrity verification;

anomaly detection;

incident-response procedures;

secure logging; and

human emergency intervention.

14. Case Law: State v. Loomis (United States)

Although State v. Loomis, 881 N.W.2d 749 (Wis. 2016), was not an energy case, it is highly relevant to algorithmic governance.

The Wisconsin Supreme Court considered the use of the COMPAS algorithm in sentencing. The court permitted consideration of the algorithm subject to important limitations and cautions. It emphasised that the algorithm could not simply become the determinative basis of the judicial decision. (Justia Law)

Relevance to energy law

The principle can be translated into energy governance:

AI may support a legally responsible decision-maker, but the existence of an algorithmic recommendation does not automatically transfer legal authority from the human institution to the algorithm.

For electricity regulators, system operators and utilities, this supports a model where AI provides evidence or recommendations while legally authorised humans retain responsibility for consequential decisions.

15. Case Law: Bates v Post Office Ltd

The English litigation concerning the Post Office's Horizon computer system provides another important lesson.

In Bates v Post Office Ltd [2019] EWHC 3408 (QB), the High Court examined serious issues concerning the reliability of the Horizon system and evidence derived from it. (Courts and Tribunals Judiciary)

The broader significance for algorithmic governance is that computer-generated information cannot automatically be treated as infallible simply because it originates from a sophisticated technological system.

The later litigation concerning the Horizon system has continued, including proceedings in 2026 involving affected sub-postmasters and allegations concerning the system's reliability. (Courts and Tribunals Judiciary)

Energy-sector lesson

A utility should not be able to argue:

"The AI/system generated the result, therefore the result must be correct."

Instead, critical AI outputs require:

validation;

human scrutiny;

evidence preservation;

technical auditing; and

mechanisms for challenging erroneous outcomes.

16. Case Law and Due Process: General Principle

The reasoning in Loomis illustrates a broader legal issue: when algorithmic tools influence consequential decisions, legal safeguards may require attention to:

transparency;

accuracy;

limitations;

independent reasoning;

non-determinative use;

reviewability; and

procedural fairness.

These concepts can be adapted to energy regulation.

For example, if an AI system recommends denying a grid connection or disconnecting a consumer, the affected party may require an appropriate opportunity for review, depending on the applicable statutory framework.

17. AI-Based Electricity Market Governance

AI can also be used in electricity markets to:

forecast prices;

identify congestion;

optimise bids;

detect market anomalies;

manage demand response; and

coordinate distributed energy resources.

This creates a regulatory question:

Should an AI-generated market decision be treated as equivalent to a human commercial decision?

A hybrid framework should distinguish between:

Automated optimisation

AI optimises an authorised strategy.

Automated execution

AI submits or executes transactions.

Human accountability

A responsible entity remains legally accountable for the system's operation.

This is particularly important where algorithms could potentially produce coordinated or anti-competitive market behaviour.

18. AI and Consumer Protection

AI-based energy systems increasingly interact directly with consumers through:

smart meters;

dynamic tariffs;

automated demand response;

home energy-management systems; and

distributed-generation platforms.

A hybrid governance model should ensure that consumers can understand important consequences of automated decisions.

For example, if an algorithm automatically modifies a consumer's demand-response participation, there should be appropriate mechanisms for:

notification;

correction;

complaint;

review; and

compensation where legally justified.

19. Emergency Situations

Emergency grid operations create a special problem.

Suppose an AI detects a rapidly developing grid instability and predicts that immediate action is necessary.

Waiting for human approval could potentially worsen the situation.

Therefore, a hybrid model may permit pre-authorised automated emergency actions.

For example:

AI detects specified instability → predefined protection protocol activates automatically → human operator is immediately notified → operator evaluates and manages the continuing event.

This creates a crucial distinction between:

pre-authorised automated action and unrestricted autonomous decision-making.

The former can operate within legally established boundaries.

20. Institutional Governance Model

A comprehensive hybrid governance structure could be organised as follows:

Level 1 – AI system

data collection;

prediction;

optimisation;

anomaly detection.

Level 2 – System operator

real-time monitoring;

approval;

intervention;

emergency management.

Level 3 – Utility management

operational policies;

cybersecurity;

AI procurement;

internal controls.

Level 4 – Independent regulator

compliance;

audits;

investigations;

consumer protection.

Level 5 – Government/legislature

statutory framework;

national energy policy;

critical-infrastructure policy.

This creates a multi-layer accountability architecture.

21. Liability in Hybrid AI Systems

One of the most difficult questions is liability.

Suppose an AI-controlled system incorrectly dispatches electricity and causes financial losses.

Potential questions include:

Was the AI properly designed?

Was the data reliable?

Did the utility follow regulatory requirements?

Did the operator ignore a warning?

Was the AI model properly tested?

Did the software developer breach contractual obligations?

Was there a cybersecurity attack?

Was human intervention technically possible?

Were regulatory safety procedures followed?

Liability should therefore be allocated according to control, duty, foreseeability and causation, rather than merely blaming "the AI."

AI itself generally should not become a convenient legal black box through which human and institutional accountability disappears.

22. Regulatory Sandboxes

Energy regulators can use regulatory sandboxes to test AI systems before large-scale deployment.

A sandbox can establish:

limited geographical deployment;

defined user population;

technical monitoring;

safety thresholds;

human oversight;

incident reporting;

model validation; and

sunset/review provisions.

This allows regulators to observe actual system behaviour without immediately exposing the entire electricity system to a new technology.

23. Certification of Energy AI Systems

A future regulatory framework could require certification based on:

Technical criteria

reliability;

accuracy;

robustness;

cybersecurity.

Governance criteria

human oversight;

auditability;

explainability;

accountability.

Legal criteria

statutory compliance;

consumer protection;

data governance;

emergency procedures.

Operational criteria

testing;

fail-safe mechanisms;

human override;

incident response.

Certification could be periodically renewed because AI models can change after deployment.

24. Continuous Monitoring

AI systems should not be considered safe merely because they passed testing at the time of deployment.

Performance can change because:

electricity demand changes;

renewable penetration increases;

weather patterns change;

consumer behaviour changes;

grid topology changes; or

new forms of cyberattack emerge.

The CEA's ongoing work on grid connectivity, technical standards and system-related regulation illustrates the dynamic nature of electricity-system regulation. (Central Electricity Authority)

Consequently, AI governance should include continuous validation and periodic reassessment.

25. Key Legal Principles for Hybrid Human–AI Energy Governance

A mature legal framework should incorporate the following principles:

1. Human accountability

AI cannot eliminate responsibility of legally authorised institutions.

2. Proportionality

Human oversight should correspond to the risk of the AI application.

3. Explainability

Important AI decisions should be sufficiently understandable for oversight.

4. Contestability

Affected parties should have appropriate mechanisms for review.

5. Auditability

Consequential decisions should be reconstructable.

6. Safety

AI must operate consistently with electricity-system safety requirements.

7. Cybersecurity

AI systems must be protected against manipulation and attack.

8. Non-discrimination

Automated systems should be tested for unjustified differential effects.

9. Reliability

AI should not be deployed without adequate technical validation.

10. Fail-safe design

Systems should have predetermined responses to AI failure.

26. Proposed Legal Model for India

India could develop a sector-specific AI Governance Framework for Electricity Systems incorporating:

AI Risk Classification → Certification → Human-Oversight Plan → Deployment → Continuous Monitoring → Incident Reporting → Independent Audit → Regulatory Review

Such a framework could be integrated with existing electricity institutions rather than creating an entirely separate regulatory structure.

The CEA already possesses statutory functions relating to technical standards, safety and grid standards under the Electricity Act, 2003. (Central Electricity Authority)

The Central Electricity Regulatory Commission and State Electricity Regulatory Commissions could address AI-related issues within their respective regulatory jurisdictions, while system operators could implement operational controls.

27. Advantages of Hybrid Governance

Hybrid governance can provide:

faster decision-making;

improved forecasting;

better renewable integration;

enhanced grid reliability;

predictive maintenance;

improved demand management;

greater operational efficiency;

human accountability; and

improved resilience.

Its principal advantage is that it attempts to combine machine-scale computational capability with human legal judgment and institutional responsibility.

28. Challenges

Nevertheless, significant challenges remain:

Accountability gap

It may be difficult to identify which actor caused an AI-related failure.

Automation bias

Human operators may accept AI recommendations without adequate scrutiny.

Black-box systems

Complex models may be difficult to explain.

Cybersecurity

Manipulated data can produce manipulated decisions.

Model drift

Performance can deteriorate over time.

Regulatory fragmentation

Electricity, AI, cybersecurity, data protection and consumer law may overlap.

Skills gap

Regulators and operators need both energy-sector and AI expertise.

29. Conclusion

Hybrid Human–AI Governance Models provide a legal and institutional framework for incorporating artificial intelligence into increasingly complex electricity systems without abandoning human responsibility.

The fundamental principle is that AI should augment institutional decision-making rather than obscure or replace legal accountability. AI can forecast demand, identify faults, optimise dispatch and detect risks, while authorised human actors retain responsibility for consequential decisions.

Indian electricity law already contains institutional foundations relevant to such governance. The Electricity Act, 2003 assigns the CEA responsibilities concerning technical standards, safety and grid standards, while Regional Power Committees perform important coordination and grid-stability functions. (Central Electricity Authority)

Cases such as State v. Loomis demonstrate the legal importance of limiting algorithmic tools so that they support rather than replace accountable decision-makers. (Justia Law) The Bates v Post Office litigation similarly illustrates why technological outputs should be subject to scrutiny rather than automatically treated as infallible evidence. (Courts and Tribunals Judiciary)

For energy systems, the emerging legal model should therefore be based on human oversight, risk-based automation, transparency, auditability, cybersecurity, contestability, technical validation and clearly allocated responsibility. The objective is not to prevent AI from governing aspects of electricity operations, but to ensure that when AI participates in governance, its authority remains embedded within a legally accountable human and institutional structure.