Artificial Intelligence and Accountability of Automated Decisions

1 Overview

The study and creation of intelligent beings that observe their surroundings and take actions that increase their chances of success is the focus of artificial intelligence (AI), a subfield of computer science. “The ability to hold two different ideas in mind at the same time and still remain the ability to function” is one definition of artificial intelligence.¹

AI is being incorporated more and more into financial services, e-commerce, banking, healthcare, education, policing, recruitment, governance, and other facets of social and economic life in India. There is now a lot of potential to improve public administration and efficiency due to the growing use of AI. Simultaneously, it has raised challenging legal issues regarding accountability for decisions made by automated systems.

When an automated judgment impacts a person’s legal rights, liberty, dignity, means of subsistence, or economic interests, the issue becomes very significant. An AI-based system might deny a loan application, flag someone as a possible fraudster, rate employment candidates, assess a person’s eligibility for a service, help with medical diagnostics, or have an impact on political decisions. The impacted individual must have a meaningful way to find out why the decision was taken and who is accountable if it is unfair or discriminatory.

The foundation of the conventional legal system is institutional or human accountability. The law establishes the repercussions when an individual conducts an act, an authority makes a decision, or a business renders a service. This structure is complicated by AI because the final result may be produced by algorithms that have been trained on enormous datasets and may involve multiple parties, such as developers, data providers, technology businesses, governmental bodies, and end users. This leads to what could be called an accountability gap.

Therefore, the crucial question is not just whether AI should be allowed to make judgments, but also whether these decisions can be integrated into India’s current legal and constitutional framework without jeopardizing fundamental rights.

An essential place to start is the Indian Constitution. Equal protection under the law and equality before the law are guaranteed under Article 14. While Article 21 safeguards life and individual liberty, Article 19 guarantees certain freedoms. These clauses have been read by Indian constitutional law in a way that prioritizes proportionality, non-arbitrariness, fairness, dignity, and privacy.

This paper’s goals are to determine whether India’s current legal system can guarantee accountability for judgments made by AI and to pinpoint the fundamental ideas needed to create an all-encompassing Indian AI accountability system.

2 Automated Decision-Making: Meaning and Nature

Systems, software, or procedures that use computation to support or replace government decisions, judgments, and/or policy implementation that affect opportunities, access, freedoms, rights, and/or safety are referred to as “automated decision systems.” Predicting, categorizing, optimizing, identifying, and/or recommending are all possible tasks for automated decision systems.²

In general, automated decision-making refers to the use of computer-based systems to analyze data and generate conclusions, suggestions, or categorizations pertaining to a person or group.

Automated processing is acknowledged under the Digital Personal Data Protection Act, 2023. According to the Act, “automated” refers to a digital process that can process data or respond to instructions automatically.³

There are many levels of automated decision-making.

1. Completely Automated Making of Decisions - Information is processed by the system and the outcome is generated without significant human involvement.

2. AI-Powered Decision-Making - AI generates a prediction or recommendation, but the ultimate decision is made by a human authority.

3. AI-Assisted Human Decision Making - AI only helps the human decision-maker by arranging data or spotting trends.

The distinction has legal implications. Ordinary legal duty can usually stay with the person using the decision-making capacity when AI only helps. However, more accountability measures are required when an organization successfully transfers decision-making to an automated system. Automation bias is a major issue. An algorithmic recommendation may be accepted by human decision-makers without independent review because they believe it to be intrinsically impartial. Therefore, having a human present at the conclusion of an automated process does not always equate to true human oversight.

2.1 Algorithmic Accountability

AI is being used more and more in hiring, worker classification, risk assessment, and other labor choices. However, accountability may get dispersed across developers, vendors, deployers, and organizations when AI makes mistakes. There should always be a clearly identified authority in charge of the result, regardless of whether an algorithm is used or outsourced.

Algorithmic responsibility requires transparency, precise data, human oversight, audits, and clear ownership. Developers should be held responsible for predictable design defects, while deployers and organizations must ensure appropriate use, accurate data, monitoring, evaluation, and effective remedies. Human officials must be able to contest or veto decisions made by AI.

AI should ultimately be viewed as a governance and compliance concern rather than just a technical instrument. Organizations should keep accurate records of their data, decision-making procedures, human evaluations, and remedial measures. The basic idea is that while AI can help with decision-making, it should never take the role of human accountability or the capacity to clarify and correct decisions.

2.2 Algorithmic Accountability and Article 14

Perhaps the most significant constitutional clause governing discriminatory or unfair automated decisions is Article 14. The ruling in E.P. Royappa⁴ by the Supreme Court held that arbitrary State action is inherently incompatible with equality. This idea was further expanded by the Court in a number of instances.

The connection to AI is simple. Because an algorithm applies the same mathematical procedure to everyone, it may appear neutral. However, equality is not always the result of mathematical neutrality. An AI hiring system that has been educated on past job data, for instance, might replicate past discriminatory trends. The AI might pick up on and replicate a pattern if prior hiring decisions consistently favored a particular group of candidates. Similarly, if the underlying data is erroneous or incomplete, an algorithm that determines a group’s eligibility for government benefits may disproportionately exclude that group. These results could give rise to Article 14 issues.

Therefore, rather than only focusing on whether discriminatory instructions were purposefully included into the algorithm, the constitutional test should consider the decision’s impact and rationale.

The concept of manifest arbitrariness is also affected by the Supreme Court’s ruling in Shayara Bano v. Union of India.⁵ The doctrine’s wider focus on arbitrariness offers a crucial conceptual foundation for analyzing automated governmental decision-making, despite the fact that it was created in the context of legislation. The simple fact that “the computer made the decision” shouldn’t make an AI system constitutionally acceptable. The decision made by the State remains an act of the State even if it uses an algorithm to wield public power.

2.3 Procedural Fairness and Article 21

Because automated judgments may have a direct impact on life, liberty, and dignity, Article 21 is equally important. The Supreme Court ruled in Maneka Gandhi v. Union of India that the process envisioned under Article 21 cannot be capricious, unjust, or irrational.⁶

AI is significantly impacted by this idea. Let’s say an automated system mistakenly determines that a person is ineligible for a government subsidy. The judgment may create major procedural-fairness issues if the person is not given an explanation or a chance to contest the outcome. Thus, the right to be heard, the audi alteram partem principle, is pertinent.

Typically, an individual impacted by a big automated decision should have:

1. Notification of the ruling

2. Details about automated processing

3. Access to pertinent personal data

4. A clear justification of the decision’s main rationale

5. A chance to rectify false information

6. A significant human evaluation

7. A grievance or appeal process.

The severity of the repercussions should determine how strong these protections are.

A recommendation made by AI regarding a person’s tastes for entertainment does not need to be protected in the same way as an automated choice that affects work, liberty, welfare benefits, healthcare, or access to basic services.

3 AI and Privacy Rights

Because contemporary AI systems rely largely on personal data, the relationship between AI and privacy is especially crucial in India. The Supreme Court unanimously acknowledged privacy as a fundamental right in Justice K.S. Puttaswamy (Retd.) v. Union of India.⁷ The ruling linked privacy to liberty, autonomy, and dignity. Informational privacy has been treated as constitutionally significant in the Court’s subsequent jurisprudence.

AI raises privacy issues in a number of ways:

• 1.
  Gathering Information – For training and operation, AI systems need a lot of data. Information privacy may be compromised by excessive collecting.
• 2.
  Data Processing - It is possible that information provided for one purpose will be handled for another.
• 3.
  Profiling - AI is able to construct profiles about people by combining various kinds of information.
• 4.
  Inference - AI has the ability to deduce sensitive traits from seemingly unremarkable data.
• 5.
  Maintaining - Additional concerns arise when personal data is stored for an extended period of time.

Therefore, the fundamental right to privacy necessitates taking into account not only what data is gathered but also how it is analyzed and what judgments are taken based on that analysis. In particular, the proportionality principle is pertinent. A restriction on privacy must be supported by the law, have a justifiable goal, be logically related to that goal, and refrain from unduly interfering with people’s rights.

3.1 The 2023 Digital Personal Data Protection Act

An essential part of India’s existing regulatory framework for data governance is the Digital Personal Data Protection Act, 2023⁸ (“DPDP Act”). The Act aims to control how digital personal data is processed while taking into account both legal processing needs and individual data protection interests.⁹ According to the Act, a Data Fiduciary is an individual who, either alone or in concert, chooses how and why to process personal data.¹⁰ This idea is especially pertinent to AI since the organization that decides how and why personal data is processed will often have significant control over an AI system.

The Act offers a number of rights that are pertinent to automated decision-making, such as the right to access personal data information, the right to have personal data corrected and erased, the right to file a grievance, and related rights and obligations for the processing of personal data.¹¹ The Act also creates the Data Protection Board of India and stipulates penalties and adjudication.¹² The Act has limited but important implications for AI. AI accountability is not the same as data protection.

Even if an AI system processes personal data in compliance with relevant data-protection regulations, it may nonetheless produce a judgment that is discriminatory or unfair. As a result, the DPDP paradigm ought to be viewed as a part of a larger framework for AI accountability. Both are necessary for a thorough system: Data accountability: Was the processing of personal data done in a responsible and legal manner? Decision accountability: Was the choice made using such information impartial, truthful, nondiscriminatory, and subject to appropriate review?

3.2 The Information Technology Act of 2000 and the Accountability of Artificial Intelligence

An essential component of India’s digital legal system is still the Information Technology Act, 2000¹³ (the “IT Act”). The IT Act’s provisions regarding electronic records, intermediary responsibilities, cybersecurity, and computer-related offenses may apply to AI-enabled activities depending on the situation, even though it was passed prior to the widespread use of modern generative AI and sophisticated machine-learning systems. Nevertheless, the IT Act is not a comprehensive law pertaining to AI accountability.

A comprehensive framework governing algorithmic bias, explainability, AI effect assessments, human monitoring, automated governmental choices, algorithmic audits, and full culpability for actions made by AI is lacking. The necessity for sector-specific and cross-sector AI governance concepts is demonstrated by this regulatory gap.

4 The “Right to Explanation” and Natural Justice

The need for natural justice is one of the most significant tenets of Indian administrative law. One of the fundamental tenets of administrative and constitutional law is the idea of natural justice, which stands for the legal system’s moral conscience. It embodies the values of justice, equity, and fairness that direct judicial and quasi-judicial processes.¹⁴ Generally speaking, a person impacted by an administrative decision should be given a fair chance to argue their position, especially if the decision negatively impacts their rights or legitimate interests. This principle is strongly related to the duty to give reasons. Three goals are served by a well-reasoned decision:

1. It shows that the authority used its judgment.

2. It enables the impacted person to comprehend the rationale for the choice.

3. It makes judicial review easier.

If the authority only claims that an AI system produced the result, automated decision-making could compromise these goals. Consequently, Indian law ought to acknowledge the notion of meaningful explanation in important circumstances. It is not necessary to reveal the entire source code in order to provide a meaningful explanation. Rather, an impacted person should typically be able to learn:

• •
  That AI was used
• •
  The AI system’s purpose
• •
  The main factors taken into account
• •
  The pertinent data relied upon
• •
  Whether a human reviewed the outcome
• •
  The system’s limitations, if applicable
• •
  The procedure for contesting the result.

This strategy safeguards legitimate interests in cybersecurity and intellectual property as well as transparency.

5 Indian Equality Law and Algorithmic Bias

When systematic mistakes in machine learning algorithms result in unjust or discriminating outcomes, this is known as algorithmic bias. Algorithmic bias is the result of AI systems reproducing and maybe amplifying social prejudices and inequities in their decision-making processes after being educated on historical data that reflects these biases.¹⁵ One of the biggest risks associated with automated decision-making is algorithmic bias.

Bias can come from:

1. Training Information – Incomplete or unrepresentative training data could lead to uneven results from the final system.

2. Discrimination Based on History – Discriminatory patterns ingrained in past decisions can be replicated by AI.

3. Proxy Variables – Indirect correlations between seemingly neutral variables and legally or constitutionally sensitive traits are possible.

4. Quality of Data – Erroneous conclusions might result from erroneous or out-of-date information.

5. Implementation – Applying a system designed for one demographic to another may result in unreliable outcomes.

In India, this problem becomes particularly important because of the constitutional commitment to equality and social justice. AI systems used by public authorities should therefore be evaluated not merely for overall accuracy but also for differential impact upon different groups. The State should be required to demonstrate that a high-risk AI system is based on rational criteria and does not produce unjustified discriminatory outcomes.

5.1 Automated Government Decisions and Proportionality

According to the proportionality doctrine, AI-based government actions that impact basic rights must have a legal foundation, a legitimate purpose, a rational connection, and necessity, while weighing the advantages to the public against the risks to individuals. As acknowledged in K.S. Puttaswamy (Retd.) v. Union of India,¹⁶ less restrictive options and privacy issues must be taken into account before adopting AI only because it is technologically feasible.

5.2 Human Supervision and Responsibility

AI-based decision-making requires meaningful human oversight. Officials must be able to obtain pertinent data, question, review, and overturn AI outputs, document justifications, and maintain legal accountability. In areas like policing, welfare, healthcare, employment, and taxation where AI judgments have a substantial impact on fundamental rights, human review should become more important.

5.3 AI Developers’ and Deployers’ Liability

Depending on variables like control, foreseeability, negligence, and causation, AI developers may be held accountable for careless design, insufficient testing, security lapses, predictable hazards, or discriminating systems. Deployers must make sure AI is appropriate for its intended use, employs correct data, is checked for discrimination, and offers human oversight and remedies. Therefore, each actor’s job and level of control should determine their level of responsibility.

5.4 Accountability of the Government

Even when private corporations build AI systems, government authorities are nonetheless accountable under the constitution. Legality, equity, nondiscrimination, privacy, proportionality, and efficient remedies must all be guaranteed. Constitutional responsibility cannot be outsourced in the same way as AI technology.

In the private sector, companies like Microsoft (Copilot) and Salesforce (Einstein) are creating AI capabilities that allow businesses to coordinate decision-making across functions at the enterprise level, while enterprise software providers like SAP and Oracle integrate finance, logistics, procurement, and human resources within unified systems. Although this integrated paradigm has not yet been fully embraced by public administration, the direction is obvious. The next phase of change will take place at the level of governance architecture rather than at the level of specific instruments.¹⁷

6 Important Judicial Decisions in India

6.1 E.P. Royappa v. State of Tamil Nadu¹⁸

The Supreme Court linked non-arbitrariness and equality. Because an algorithmically generated judgment might still be arbitrary, the principle is applicable to automated decision-making. Technology cannot change a state’s arbitrary actions into ones that are lawfully permissible.

6.2 Maneka Gandhi v. Union of India¹⁹

The Court mandated fairness and rationality in processes impacting individual liberty and rejected a limited interpretation of Article 21 (the connection between Articles 14, 19, and 21). The case backs up the idea that meaningful procedural protections should be provided to anyone impacted by important automated government decisions.

6.3 Justice K.S. Puttaswamy (Retd.) v. Union of India²⁰

The nine-judge bench acknowledged privacy as a basic right. Because automated systems rely on the gathering, processing, analysis, and inference of personal data, the ruling is especially pertinent to artificial intelligence. The Court acknowledged the intimate relationship between privacy and autonomy and dignity.

6.4 Shayara Bano v. Union of India²¹

A helpful framework for analyzing irrational or arbitrary State behavior is provided by the Supreme Court’s consideration of manifest arbitrariness. Similar worries can be raised by an automated decision that yields results that are illogical or irrational, especially when basic rights are at stake.

6.5 Anuradha Bhasin v. Union of India²²

The Supreme Court focused on legality, reasonableness, and proportionality while examining limitations on basic freedoms in relation to internet restrictions.

6.6 Internet & Mobile Association of India v. Reserve Bank of India²³

The Supreme Court examined regulatory limitations pertaining to virtual currencies in applying proportionality standards. The case illustrates the need for constitutional principles to strike a balance between technological progress and regulatory authority.

7 Developments in Indian Regulation Regarding AI

Instead of having a single comprehensive AI law, India has historically adopted a technology-friendly approach to AI, depending on innovation, responsible AI, and sector-specific regulation. Constitutional law, data protection, information technology laws, administrative law, judicial review, and government policies are all included in its regulatory framework. The government is putting more emphasis on responsible, secure, and reliable AI governance, especially through MeitY. Instead of depending only on current general rules, India requires a clear accountability framework that guaranties efficient oversight as the usage of AI grows.

Under the IndiaAI Mission, MeitY published the India AI Governance Guidelines in November 2025. Seven fundamental ideas—often referred to as “sutras”—form the basis of the guidelines:

• 1.
  Trust as the foundation
• 2.
  A human-centric (people-first) perspective
• 3.
  Responsible innovation, or innovation over restraint
• 4.
  Justice, equity, and nondiscrimination
• 5.
  Accountability
• 6.
  Transparency and comprehensibility based on design
• 7.
  Security, safety, and sustainability²⁴

8 AI Accountability: Proposed Indian Framework

India ought to think about implementing a thorough, risk-based framework built on the following tenets.

1. Required AI Impact Evaluation - Before being deployed, high-risk AI systems should go through an impact assessment. The evaluation ought to take into account confidentiality, equality, prejudice, precision, safety, human rights, possible harm to society, and whether remedies are available.

2. Algorithmic Audits - For high-risk systems, independent audits ought to be carried out on a regular basis. Both technical performance and legal compliance should be looked at during the audit.

3. Required Human Evaluation - When a choice has a significant impact on basic rights, AI shouldn’t have the last say.

4. The right to a meaningful explanation - Reasonable explanations for important automated decisions should be provided to affected parties.

5. The Right to Contest - A person ought to be able to contest a decision made by AI in front of a separate human authority.

6. Accuracy of Data - When making important decisions, organizations should confirm the relevance and quality of the data.

7. Testing for Algorithmic Equality - It is important to examine high-risk systems for discriminating results.

8. Audit Trails - Important AI systems ought to save enough documentation to reconstruct the decision-making process.

9. Responsibility Distribution - The obligations of developers, deployers, and data providers should be specified in contracts. However, statutory or constitutional requirements should not be eliminated by contractual agreements.

10. Remedies and Compensation - Effective remedies should be available to those who suffer legally recognized injury as a result of unlawful or careless AI deployment.

9 Difficulties in India with AI Accountability

A number of issues need to be resolved.

• •
  Absence of an All-Inclusive AI-Specific Law - India currently heavily depends on current laws and emerging regulatory frameworks.
• •
  Complexity of Technology - It’s possible that judges, attorneys, and regulators lack the specialized technical expertise needed to assess complicated AI systems.
• •
  Trade Secrets - Businesses may object to algorithm publication on the grounds of intellectual property protection.
• •
  Insufficiently Standardized Auditing - For algorithmic testing and certification, uniform criteria are required.
• •
  Dispersed Accountability - Developers, data producers, cloud service providers, and deployers may be involved in AI systems.
• •
  Quick Changes in Technology - As AI technology advances, laws may become out of date.
• •
  Harmonizing Regulation and Innovation – While under-regulation may expose people to grave rights breaches, over-regulation may deter technical advancement.

These difficulties show why India should embrace adaptable values as opposed to strict technology regulations.

9.1 The Necessity of a Human-Centered Strategy

AI governance must guarantee that technology upholds constitutional principles and does not take the role of human accountability. AI can help decision-makers, but it cannot be used to evade responsibility. People who are impacted by AI choices need to be able to recognize the responsible authority, comprehend the rationale behind the decision, and request a meaningful review. Therefore, in order to prevent technological efficiency from superseding fundamental rights, AI governance in India must safeguard equality, liberty, dignity, and the rule of law.

10 Conclusion

By enhancing productivity, data analysis, and governance, artificial intelligence is revolutionizing decision-making in India. However, automated decisions may also be opaque, discriminating, erroneous, and challenging, raising significant accountability issues. Important protections are provided by the Indian Constitution, especially Articles 14, 19, and 21, natural justice, judicial review, and the right to privacy acknowledged in Puttaswamy. Although data privacy is further strengthened by the Digital Personal Data Protection Act, 2023, a more comprehensive framework for automated decision-making is required. Transparency, explainability, human oversight, nondiscrimination, audits, and efficient remedies should all be guaranteed by such a system. In the end, AI should continue to be governed by human accountability and the rule of law, guaranteeing that technology advancement fosters efficiency without jeopardizing equality, privacy, dignity, or fundamental rights.