1. Introduction

Artificial intelligence (AI) is increasingly being used in electricity and commodity markets to forecast demand, predict prices, determine bidding strategies, optimise portfolios, execute trades, and respond automatically to changing market conditions. An AI system may therefore generate a market action—for example, submitting a bid, cancelling an order, changing a trading position, or responding to congestion—without a human approving each individual transaction.

This creates a difficult legal question: who is liable when an AI-generated market action causes financial loss, market manipulation, regulatory violations, or disruption of the electricity market?

The central legal principle emerging from existing law is that the autonomous nature of an algorithm does not normally make the algorithm itself legally responsible. Liability generally remains with the human or legal entity that designed, deployed, controlled, authorised, or failed adequately to supervise the system.

Importantly, there are still relatively few reported decisions specifically concerning generative AI making electricity-market decisions. Consequently, existing cases involving algorithmic trading, automated order systems, market manipulation, and electricity-market regulation provide the principal legal analogies.

2. Meaning of AI-Generated Market Actions

An AI-generated market action occurs when an AI or machine-learning system materially determines a market participant's conduct.

Examples include:

• submitting electricity bids automatically;
• modifying bids in response to price signals;
• withdrawing or cancelling bids;
• purchasing or selling electricity;
• participating in balancing or ancillary-service markets;
• forecasting congestion and changing dispatch;
• automatically responding to transmission constraints;
• participating in demand-response programmes;
• determining quantities and prices;
• identifying arbitrage opportunities;
• managing battery storage;
• coordinating distributed energy resources.

EU energy-market law already recognises algorithmic trading in wholesale energy products. Under the EU REMIT framework, algorithmic trading includes trading where a computer algorithm automatically determines parameters such as whether to initiate an order, its timing, price or quantity, with limited or no human intervention. EUR-Lex

AI therefore does not necessarily create an entirely new category of legal conduct. Rather, it makes existing automated market conduct more complex because modern AI can learn, adapt and generate strategies that were not explicitly programmed line-by-line.

3. The Basic Principle of Liability

A useful framework is:

AI may generate the action, but legal liability ordinarily follows the person or entity legally responsible for putting the AI into the market.

Potentially responsible actors include:

1. AI developer
2. Software supplier
3. Electricity trader
4. Generator
5. Power-market participant
6. Aggregator
7. Broker
8. Power exchange
9. System operator
10. Data provider
11. AI deployer/operator

Liability depends on the particular relationship between these parties.

For example, suppose an electricity trader deploys an AI bidding system and the system accidentally submits bids ten times larger than intended. The trader may potentially face regulatory consequences even though no human employee manually entered the erroneous bids.

The crucial questions become:

• Who deployed the system?
• Who controlled it?
• Who benefited from the market action?
• Who had the duty to supervise it?
• Were adequate safeguards implemented?
• Was the system properly tested?
• Was the AI acting within its authorised parameters?
• Did the operator know or reasonably foresee the risk?
• Did the operator respond appropriately after detecting the problem?

4. Liability for Intentional AI Market Manipulation

The most straightforward situation is where AI is deliberately used to manipulate a market.

Suppose a trader intentionally instructs an AI system to:

• place misleading orders;
• create artificial demand;
• generate false liquidity;
• manipulate prices;
• engage in spoofing;
• rapidly cancel orders;
• create artificial congestion;
• distort electricity prices.

The fact that the AI performs the conduct automatically does not normally eliminate the user's responsibility.

Case: United States v. Coscia

One of the most important cases demonstrating this principle is United States v. Coscia.

Michael Coscia used computer algorithms to conduct spoofing in futures markets. The algorithms placed large orders designed to create a false impression of market interest while the trader intended to cancel those orders before execution.

The jury convicted Coscia of commodities fraud and spoofing. The conviction was subsequently upheld on appeal. The evidence showed that the algorithms had been specifically designed to execute the strategy requested by Coscia. Department of Justice

The legal significance is substantial:

The use of an automated algorithm does not shield the person who intentionally designed or commissioned the unlawful strategy.

This principle would be highly relevant to AI-generated electricity-market manipulation.

5. CFTC v. Panther Energy Trading

The regulatory proceedings involving Panther Energy Trading LLC and Michael Coscia provide another important example.

The CFTC found that Panther and Coscia used a computer algorithm designed to rapidly place and cancel orders in futures markets. The conduct involved numerous commodity contracts, including energy-related contracts. The CFTC imposed monetary penalties and trading restrictions. Commodity Futures Trading Commission

This case establishes an important distinction:

Lawful algorithmic trading

An algorithm may legitimately:

• optimise execution;
• react to price changes;
• hedge risk;
• respond to supply and demand;
• manage a portfolio.

Unlawful algorithmic trading

The algorithm cannot lawfully be used to:

• create false market signals;
• manipulate prices;
• spoof;
• deceive other participants;
• artificially influence supply or demand.

The fact that the algorithm operates at machine speed does not change the underlying legal character of the conduct.

6. Liability for AI Errors Rather Than Manipulation

The more difficult problem arises when the AI has no intention to manipulate the market.

Consider an AI system that unexpectedly produces:

electricity purchase order = 500 MW

when the trader intended:

electricity purchase order = 50 MW.

There may be no fraudulent intention.

Nevertheless, the resulting transaction could:

• distort market prices;
• create financial losses;
• cause congestion;
• affect other participants;
• create imbalance;
• trigger regulatory intervention.

The question then becomes one of negligence, supervision, system controls and regulatory responsibility.

7. Knight Capital: The Leading Automated-Trading Error Example

A particularly important analogy is In the Matter of Knight Capital Americas LLC.

In 2012, Knight Capital's automated trading system malfunctioned following a software deployment problem. Instead of processing a relatively small number of customer orders, the system generated millions of erroneous orders.

The SEC found that Knight's system sent more than four million orders into the market during approximately 45 minutes, resulting in more than 397 million shares being traded and billions of dollars in unintended positions. Knight ultimately suffered losses exceeding $460 million. SEC

The SEC imposed a $12 million penalty for violations of the market-access rule.

The important legal principle is not that Knight's algorithm was "guilty." Rather, the liability focused on the regulated entity's failure to maintain adequate controls over automated market access.

The SEC identified failures involving:

• inadequate pre-trade controls;
• inadequate financial risk controls;
• inadequate software testing;
• inadequate code deployment procedures;
• insufficient supervisory procedures;
• inadequate response to technological problems;
• inadequate review of automated trading risks. SEC

Relevance to AI

This is highly relevant to AI-generated market actions.

If an AI system produces an unexpected market action, the operator may be asked:

Why did you permit an AI system to access the market without sufficient safeguards?

Thus, liability may arise from failure to control the AI, even where the particular market action was not deliberately programmed.

8. AI "Black Box" Does Not Automatically Remove Responsibility

Modern machine-learning systems can be difficult to explain.

An AI may produce a particular bid because of:

• thousands of historical variables;
• real-time market information;
• weather forecasts;
• demand predictions;
• network conditions;
• learned correlations;
• reinforcement-learning strategies.

The operator may therefore argue:

"We cannot explain exactly why the AI generated this particular bid."

That argument does not automatically eliminate regulatory responsibility.

For regulated market participants, the more important question may be whether they had:

• appropriate governance;
• testing;
• validation;
• monitoring;
• human oversight;
• audit trails;
• risk limits;
• emergency shutdown mechanisms.

The Knight Capital proceedings illustrate why automated-system controls matter independently of whether the underlying mistake was intentional. SEC

9. Electricity-Market Context

The issue becomes particularly important in electricity markets because electricity cannot easily be stored at scale across the system and supply and demand must remain balanced.

An AI-generated market action can therefore have consequences beyond the trader's private loss.

For example:

AI forecast → automated bid → market clearing → dispatch → congestion → imbalance → system impact

A mistaken AI decision could therefore produce:

• imbalance charges;
• congestion;
• price distortion;
• transmission constraints;
• unnecessary dispatch;
• increased balancing costs;
• financial losses to other market participants;
• potentially wider reliability consequences.

This makes electricity-market AI different from an ordinary commercial AI application.

10. Indian Electricity-Market Position

In India, the legal framework must be considered principally through the Electricity Act, 2003, CERC regulations, power-market regulations, exchange rules and applicable contractual arrangements.

CERC's regulatory framework already recognises the importance of software and algorithmic systems used by power exchanges.

For example, CERC regulations require algorithms used for price discovery and market splitting by power exchanges to comply with specified requirements and provide for periodic algorithm audits. CERC also retains the ability to conduct or commission audits of such software applications. CERC India

This is significant because it demonstrates a regulatory philosophy of:

algorithmic market activity + auditability + institutional responsibility.

Therefore, if AI is integrated into electricity-market bidding or price discovery, questions of algorithm validation and auditability are likely to become increasingly important.

CERC's current regulatory framework also includes rules governing real-time electricity markets and other market mechanisms. CERC India

11. Possible Categories of Liability

A. Regulatory liability

A market participant may violate:

• electricity-market regulations;
• trading rules;
• market-access requirements;
• market-manipulation provisions;
• exchange rules;
• balancing rules.

Regulatory liability may exist even without conventional civil negligence.

B. Civil liability

Another participant may suffer losses because of an AI-generated action.

Potential claims may arise from:

• contract;
• negligence;
• misrepresentation;
• breach of statutory duty;
• market rules;
• indemnification provisions.

For example, if an AI-controlled trading system violates a contractual bidding obligation and causes another participant measurable losses, contractual liability may become relevant.

C. Administrative penalties

Regulators may impose:

• monetary penalties;
• trading restrictions;
• licence consequences;
• compliance orders;
• corrective measures;
• suspension or other regulatory sanctions.

Knight Capital illustrates how inadequate controls around automated trading can lead to regulatory sanctions. SEC

D. Criminal liability

Criminal liability becomes particularly relevant where AI is deliberately deployed to facilitate:

• fraud;
• spoofing;
• manipulation;
• false reporting;
• conspiracy;
• deliberate market abuse.

The Coscia prosecution demonstrates that a trader can be criminally responsible where algorithms are deliberately used to implement unlawful trading strategies. Department of Justice

12. Developer Versus Operator Liability

One of the most difficult questions is whether liability should fall on the AI developer or the market participant using the AI.

Example

An electricity company purchases an AI trading system from a technology company.

The AI subsequently generates unlawful bids.

Possible responsibility may be divided as follows:

The appropriate allocation depends heavily on the facts and applicable legislation.

A developer ordinarily should not automatically become liable merely because its software was used improperly. Conversely, a developer could face liability where the software itself was defectively designed, inadequately secured, deceptively marketed, or contractually warranted to perform particular functions.

13. EU Approach to AI Liability

European law illustrates an emerging approach to AI accountability.

The EU AI Act establishes obligations concerning high-risk AI systems, including requirements relating to risk management, documentation, monitoring and human oversight in applicable circumstances. EUR-Lex

Earlier EU policy work on AI liability also recognised that AI's complexity, opacity, connectivity, capacity for modification and the involvement of multiple actors create difficulties for traditional liability systems. EUR-Lex

For electricity-market AI, these principles support a regulatory model based upon:

risk assessment → documentation → monitoring → human oversight → accountability.

However, the precise application of the AI Act to a particular electricity-market system must be determined from the system's function and the relevant legal classification; one should not automatically assume that every electricity-market AI application is a "high-risk AI system."

14. The Problem of Causation

AI-generated market actions create a complicated causation problem.

Suppose:

AI error → incorrect bid → price change → competitor loss

The injured party must potentially establish:

1. the AI generated the action;
2. the action was unlawful or otherwise legally attributable;
3. the action caused the relevant market effect;
4. the claimant suffered legally recoverable damage;
5. the defendant owed the relevant duty.

Electricity markets make this especially complicated because prices can change for many simultaneous reasons:

• weather;
• demand;
• generation outages;
• fuel prices;
• transmission congestion;
• renewable generation;
• interconnector flows;
• other market participants.

Consequently, proving that a particular AI-generated action caused a particular financial loss can be technically difficult.

15. AI Hallucination and Market Actions

Generative AI creates an additional risk that conventional algorithmic trading did not always present.

A generative system could theoretically:

• misunderstand market instructions;
• generate incorrect numerical values;
• misinterpret regulatory information;
• use stale information;
• produce an internally inconsistent trading strategy;
• act on erroneous assumptions.

For that reason, generative AI should generally not be treated as an unsupervised market actor merely because it is technically capable of producing orders.

A legal compliance architecture should separate:

AI recommendation → validation → risk controls → authorised execution.

The more significant the market consequences, the greater the justification for independent validation and execution controls.

16. Attribution: Who "Made" the Decision?

A fundamental legal question is whether an AI-generated decision should legally be attributed to the company operating it.

A practical approach is to distinguish three stages:

Stage 1 — Design

Who designed the AI strategy?

Stage 2 — Deployment

Who authorised the AI to participate in the electricity market?

Stage 3 — Execution

Who allowed the resulting instruction to reach the exchange?

This creates an attribution chain:

Developer → Operator → AI → Market → Consequence

Legal liability can potentially arise at different points in this chain depending on the applicable duty.

17. The Importance of Human Oversight

Human oversight is particularly important where AI has:

• direct market access;
• authority to determine bid prices;
• authority to determine quantities;
• access to substantial capital;
• ability to cancel orders;
• ability to influence system conditions.

An appropriate control system could include:

1. maximum bid limits;
2. maximum quantity limits;
3. price collars;
4. anomaly detection;
5. human approval for unusual transactions;
6. automatic kill switches;
7. independent testing;
8. continuous monitoring;
9. audit logs;
10. model validation;
11. cybersecurity controls;
12. periodic algorithmic audits.

The lesson from Knight Capital is particularly clear: automated systems require controls designed specifically for the risks created by automation. SEC

18. Market Manipulation and AI Intent

One of the hardest issues will be determining intent.

Suppose an AI discovers that repeatedly placing and cancelling orders increases prices.

Did the operator:

• intentionally instruct the AI to manipulate prices?
• negligently allow the AI to discover the strategy?
• fail to monitor the AI?
• genuinely have no knowledge of the strategy?

The Coscia case demonstrates why evidence concerning the design and operation of the algorithm can become important in determining intent. Commodity Futures Trading Commission

For AI systems, regulators may therefore examine:

• training objectives;
• reward functions;
• prompts;
• configuration;
• system instructions;
• source code;
• model outputs;
• logs;
• human approvals;
• risk controls;
• communications concerning deployment.

19. Evidentiary Challenges

AI-generated market actions create substantial evidentiary questions.

A regulator or claimant may need access to:

• model version;
• training data;
• input data;
• output logs;
• prompts;
• system instructions;
• timestamps;
• market conditions;
• execution records;
• human interventions;
• model updates.

This makes AI audit trails extremely important.

Without reliable logs, it may become difficult to determine whether an action resulted from:

• the AI;
• human intervention;
• defective data;
• software malfunction;
• exchange infrastructure;
• cyberattack.

20. Case-Law Principles

The major cases can be synthesised as follows:

These are not all electricity-market cases, but they provide important legal principles for determining responsibility for automated market behaviour.

21. Athena Capital Research

The SEC's action against Athena Capital Research is particularly relevant to AI-driven market actions.

The SEC alleged that Athena used an algorithm called "Gravy" to conduct rapid trading near market close and manipulate closing prices. The SEC described the case as its first high-frequency-trading manipulation case. SEC

The principle is straightforward:

The sophistication and speed of an automated trading system do not transform prohibited manipulation into lawful market activity.

For electricity markets, the same reasoning could potentially apply where an AI system is deliberately configured to manipulate prices or market signals.

22. Emerging Doctrine: Responsibility Without Direct Human Execution

The developing legal approach can therefore be represented as:

Human does not need to press the button personally.

Instead, liability can arise where the human or organisation:

• authorised the system;
• designed the strategy;
• controlled market access;
• failed to establish safeguards;
• failed to monitor the system;
• ignored warning signals;
• benefited from unlawful activity.

Knight Capital demonstrates this particularly well: the legal problem was not that an employee manually entered every erroneous order, but that the firm lacked adequate controls over its automated market-access system. SEC

23. Proposed Liability Framework for AI Electricity Markets

A future electricity-market liability framework could adopt a tiered responsibility model.

Tier 1 — Intentional misuse

Where a participant intentionally uses AI for manipulation:

Primary liability → market participant/operator.

Tier 2 — Negligent deployment

Where AI causes harm because the operator failed to test or supervise it:

Primary liability → deploying entity.

Tier 3 — Defective AI technology

Where the system contains a genuine software defect attributable to its supplier:

Potential liability → developer/supplier, subject to applicable contract and product-liability law.

Tier 4 — Faulty data

Where erroneous external data causes the AI to act improperly:

Potential liability → data provider and/or operator, depending on contractual and regulatory duties.

Tier 5 — Exchange/system failure

Where the market infrastructure itself causes the error:

Potential responsibility → exchange/system operator, subject to its statutory and contractual obligations.

This model prevents the simplistic conclusion that "AI is responsible."

24. Key Legal Challenges

Several unresolved issues remain.

1. Autonomous decision-making

How much autonomy can an AI possess before conventional human attribution becomes difficult?

2. Explainability

Can a regulator establish liability when even the operator cannot explain the precise reasoning of the model?

3. Causation

How can losses caused by an AI be distinguished from ordinary market volatility?

4. Shared responsibility

How should liability be divided between developers, traders, data providers and exchanges?

5. Contractual allocation

Can contracts transfer AI-related risks between the participants?

6. Regulatory standards

What constitutes reasonable AI supervision in an electricity market?

7. Dynamic learning

If an AI changes its strategy after deployment, who is responsible for behaviour that the developer did not specifically anticipate?

25. Conclusion

Liability for AI-generated market actions should not be based simply on the proposition that "the AI made the decision." Existing algorithmic-trading jurisprudence indicates that legal responsibility generally remains connected to the human or legal entity that designs, deploys, authorises, controls or fails adequately to supervise the automated system.

The cases of Coscia, Panther Energy Trading, Knight Capital and Athena Capital Research demonstrate four particularly important principles:

1. Automation does not provide immunity from market-abuse laws.
2. Intentional algorithmic manipulation can produce direct regulatory or criminal liability.
3. A market participant may be liable for inadequate controls over an automated system even where the harmful action was an error.
4. Effective testing, monitoring, auditability and intervention mechanisms are becoming central components of responsible automated trading.

For electricity markets, these principles are especially important because an AI-generated market action can affect not merely a private trading position but also prices, congestion, balancing, dispatch and system reliability. Indian regulation already demonstrates an emphasis on auditing algorithms used for important power-exchange functions, while international regulatory experience provides increasingly strong evidence that responsibility cannot simply be transferred to the software itself. CERC India

Accordingly, the emerging legal model can be expressed as:

AI autonomy does not equal legal autonomy. The party that places AI into a regulated electricity market must ordinarily remain accountable for the risks created by that deployment, subject to the specific duties, contracts, statutes and regulatory rules applicable to each participant.