Artificial Intelligence and Law: A Critical Analysis

Introduction:

As a practicing lawyer in India, I find myself navigating uncharted legal territory almost daily. Artificial Intelligence has moved from science fiction to courtroom reality with breathtaking speed, and our legal system – built on centuries of precedent rooted in human agency, intent, and accountability – now confronts questions that our lawmakers could not have imagined even a decade ago.

When I began my legal practice, disputes centered on contracts between human parties, torts committed by identifiable actors, and intellectual property created by human minds. Today, I advise clients on liability for decisions made by algorithms, defend against allegations of copyright infringement by training datasets, and draft contracts governing AI systems whose behaviour cannot be fully predicted or controlled.

The legal challenges posed by AI are not abstract academic debates – they are immediate, practical concerns affecting my clients’ businesses, livelihoods, and fundamental rights. This article examines India’s AI landscape through the lens of a legal practitioner, analysing the regulatory frameworks we navigate, the gaps that create risk and uncertainty, and the evolving jurisprudence that will shape India’s AI-driven future.

The Current Legal Landscape: A Patchwork Approach

The Absence of AI-Specific Legislation

The first question clients invariably ask is: “What does Indian law say about AI?” The honest answer creates anxiety: India currently has no comprehensive AI-specific legislation.

Instead, we navigate a fragmented legal landscape comprising:

• Information Technology Act, 2000: Our digital law workhorse, drafted before smartphones existed
• Digital Personal Data Protection Act, 2023: Recently enacted but still being interpreted
• Copyright Act, 1957: Last substantially amended in 1994, decades before generative AI
• Consumer Protection Act, 2019: Applicable but not designed for algorithmic harms
• Indian Contract Act, 1872: Victorian-era legislation applied to 21st-century AI contracts

As lawyers, we’re essentially retrofitting 19^th and 20^th century legal frameworks onto 21^st century technology – a process that involves creative interpretation, analogical reasoning, and considerable legal uncertainty.

The India AI Governance Guidelines (November 2025)

In November 2025, the Ministry of Electronics and Information Technology released the India AI Governance Guidelines, which I initially greeted with cautious optimism. Finally, I thought, a framework specifically addressing AI governance.

However, the Guidelines are explicitly non-binding and voluntary – forming what the government terms a “techno-legal governance architecture.” While they establish seven guiding principles (trust, people-first, innovation, fairness, accountability, transparency, and safety), they create no enforceable legal obligations.

From a lawyer’s perspective, this presents challenges:

Uncertainty in Compliance: Clients ask whether compliance with the Guidelines provides legal protection. The answer is unclear. Will courts view compliance as evidence of due diligence? Will regulators consider the Guidelines when evaluating AI systems? We don’t yet know.

Lack of Enforcement Mechanisms: The Guidelines recommend practices but impose no penalties for non-compliance. This creates competitive disadvantages for responsible actors who invest in compliance while competitors ignore the Guidelines without consequence.

Evidentiary Value: In future litigation, the Guidelines may be referenced as industry standards or best practices, but their precise legal weight remains uncertain. This makes risk assessment and legal counselling challenging.

That said, I counsel clients to treat the Guidelines as a roadmap for responsible AI development. Courts and regulators will likely reference these principles when evaluating AI-related disputes, and proactive compliance demonstrates good faith – valuable in any legal proceeding.

The Proposed Draft AI Regulation Bill (2025)

The Draft AI Regulation Bill released for public consultation in early 2025 represents India’s first attempt at comprehensive AI legislation. As someone who reviewed the draft extensively and submitted detailed comments, I see both promise and concerns.

Key Provisions from a Legal Perspective:

Risk-Based Classification: The Bill categorizes AI systems as high-risk (operating in finance, healthcare, law enforcement, public safety) or low-risk. This risk-based approach is sensible and mirrors the EU AI Act, but the definitions lack precision. How do we determine which systems qualify as “high-risk”? The legal standards remain vague.

Mandatory Registration: All significant AI systems must be registered with a central regulatory authority, including technical disclosures. This creates substantial compliance burdens and raises concerns about:

• Trade secret protection: How much technical detail must be disclosed? Will this disclosure compromise proprietary information?
• Regulatory capacity: Does India have the institutional capacity to review and approve potentially thousands of AI system registrations?
• Timelines: What approval timelines apply? Can businesses deploy AI while awaiting registration?

Regulatory Sandbox: The proposed sandbox for startups to innovate without premature penalties is excellent in theory. However, the Bill lacks detail on:

• Eligibility criteria: Which entities qualify for sandbox participation?
• Scope of legal immunity: What protections does sandbox participation provide?
• Duration and exit pathways: How long can entities operate in the sandbox?

Liability Frameworks: The Bill remains notably silent on liability allocation when AI systems cause harm-perhaps the most critical legal question my clients face.

The Bill is expected to be finalized by late 2025 or early 2026, but until then, businesses operate in regulatory limbo-a state that makes legal counseling particularly challenging.

The Authorship Problem: Who Owns AI-Generated Works?

Section 2(d)(vi) of the Copyright Act, 1957, defines the author of computer-generated works as “the person who causes the work to be created.” This provision, drafted in an era of simple computer-aided design, now applies to autonomous generative AI systems – with legally absurd results.

Consider these scenarios I regularly encounter:

Scenario 1: A marketing executive uses ChatGPT to draft advertising copy. The executive provides a brief prompt; the AI generates creative content. Who is the author?

• The executive who provided the prompt?
• OpenAI, which developed the AI system?
• The countless content creators whose copyrighted works trained the model?

Scenario 2: An AI system autonomously generates thousands of images daily without human prompts. Can these works be copyrighted? If so, by whom?

Scenario 3: A researcher uses AI to analyse data and generate insights. The AI produces novel conclusions the researcher didn’t anticipate. Who owns these insights?

Indian courts have not addressed these questions directly. The Delhi High Court’s recent decision in Asian News International v. OpenAI (discussed below) touches these issues but doesn’t resolve them.

My legal advice to clients:

Document everything: Maintain detailed records of human contributions to AI-generated works – prompts, selection criteria, edits, creative direction. This documentation may establish sufficient human authorship for copyright protection.

Contractual clarity: AI development and licensing agreements must explicitly address ownership of AI-generated outputs. Don’t rely on default legal rules that may not exist or favor your position.

Register works cautiously: When registering AI-assisted works with the Copyright Office, clearly describe human contributions. Misrepresentation risks invalidating copyright.

Assume uncertainty: Until courts or legislation provide clarity, assume that AI-generated works may have limited or no copyright protection. Structure your business models accordingly.

The Training Data Dilemma: Copyright Infringement at Scale?

Perhaps the most commercially significant copyright question concerns AI training datasets. Generative AI models are trained on massive datasets – often billions of documents, images, videos, and other copyrighted works scraped from the internet without permission or compensation to copyright holders.

Is this copyright infringement? The honest answer is: we don’t know with certainty under Indian law.

The Landmark Case: Asian News International v. OpenAI (2024)

This case, currently pending before the Delhi High Court, represents the first Indian litigation addressing AI training data copyright issues. I’m following it closely as it will establish crucial precedents.

Facts: ANI, a major news agency, alleges OpenAI used its copyrighted news content – articles, photographs, videos-to train ChatGPT without permission or payment. ChatGPT then generates outputs that:

• Reproduce ANI’s content without attribution
• Produce false or misleading information attributed to ANI
• Compete with ANI’s services by providing news summaries

ANI’s Legal Arguments (which I find compelling):

Unauthorized Reproduction: Training AI requires copying content into databases-an act of reproduction requiring copyright permission under Section 14 of the Copyright Act.

Commercial Exploitation: OpenAI uses ANI’s copyrighted works to develop a commercial product (ChatGPT) and generates revenue without compensating copyright holders.

Moral Rights Violation: False attributions and misleading outputs violate ANI’s moral rights under Section 57 to protect the integrity of its works.

OpenAI’s Defense (which raises important questions):

Fair Use/Fair Dealing: Section 52 of the Copyright Act permits “fair dealing” for purposes including research and private study. OpenAI argues AI training qualifies as transformative research use.

No Substantial Reproduction: While training involves copying, the AI model itself doesn’t contain reproduced works-it contains mathematical patterns derived from those works.

Jurisdictional Challenge: OpenAI’s servers and operations are located outside India; therefore, OpenAI argues it’s not subject to Indian jurisdiction.

My Analysis: This case will be decided on multiple grounds, but I expect the Delhi High Court to:

1. Reject the jurisdictional challenge: Indian courts have long-arm jurisdiction over entities whose activities affect India, even if physically located abroad. OpenAI services Indian users; therefore, Indian courts have jurisdiction.

2. Find some copyright infringement: Training AI requires copying copyrighted works-at least temporarily-which constitutes reproduction. However, the court may recognize a limited fair dealing exception for transformative AI training.

3. Address moral rights violations: False attribution and misleading content violate authors’ moral rights-a stronger claim than pure reproduction.

4. Order limited remedies: Given AI’s societal benefits, I expect the court to balance copyright protection with technological progress, possibly ordering:

• • Attribution and licensing requirements rather than blanket prohibitions
  • Compensation mechanisms for copyright holders
  • Technical measures to prevent false attributions

Implications for Legal Practice: This case will establish whether Indian law permits training AI on copyrighted works without permission. The judgment will determine whether thousands of AI systems currently operating in India violate copyright law-a question that keeps my clients awake at night.

Text and Data Mining: The Missing Exception

The Copyright Act, Section 52, lists “fair dealing” exceptions permitting limited use of copyrighted works without permission. However, unlike the EU’s Digital Single Market Directive or Japan’s liberal copyright exceptions, Indian law lacks explicit text and data mining (TDM) provisions.

The Legal Uncertainty: Does analysing copyrighted text and data to train AI constitute:

• Fair dealing (permitted)?
• Transformative use (permitted under evolving jurisprudence)?
• Copyright infringement (prohibited)?

Indian courts haven’t decided, and until they do, AI developers face legal risk.

Government Position: Recently, in response to a parliamentary question, the Government of India clarified that “appropriate permissions need to be obtained from intellectual property rights holders before using copyright-protected work for training AI models.”

This statement indicates governmental intent but is not binding legislation. Parliamentary statements influence policy but don’t create enforceable law until codified in statutes.

Practical Implications: AI developers must either:

1. Obtain licenses from copyright holders (impractical given dataset scale)

2. Use public domain or licensed datasets (limiting AI capabilities)

3. Proceed with legal risk (hoping for eventual favorable legislation or judicial interpretation)

4. Develop AI systems outside India (creating competitive disadvantages for Indian AI)

My Recommendation to the Legislature: India should enact explicit TDM exceptions similar to EU or Japanese frameworks, balancing copyright holders’ rights with AI innovation. The current uncertainty discourages AI development and creates unnecessary litigation.

Licensing and Collective Management

The absence of established licensing or collective management frameworks for AI training creates practical obstacles.

The Problem: Individual licensing is impractical. An AI model might train on billions of documents from millions of copyright holders. Negotiating individual licenses is impossible.

Global Developments: Companies like OpenAI are entering licensing agreements with major publishers (New York Times, Associated Press, Reuters) and media houses. However, these cover only a fraction of training data.

India’s Opportunity: India should establish collective licensing organizations specifically for AI training, similar to music licensing organizations (PPL, IPRS). Copyright holders could register works; AI developers could pay licensing fees; organizations could distribute revenues.

This would:

• Compensate copyright holders fairly
• Enable AI developers to train models legally
• Provide legal certainty reducing litigation risk
• Foster innovation while respecting intellectual property

Until such frameworks exist, the copyright-AI conflict will generate extensive litigation-beneficial for lawyers like me, but inefficient for society.

Liability and Accountability: Who Answers When AI Goes Wrong?

Perhaps the most fundamental legal question I confront is: When AI causes harm, who is legally responsible?

Traditional tort law assumes human actors making intentional or negligent decisions. AI disrupts these assumptions. AI systems are:

• Autonomous: Making decisions without direct human control
• Opaque: Producing outputs through processes even developers don’t fully understand
• Adaptive: Learning and evolving after deployment
• Distributed: Involving multiple parties-developers, deployers, data providers, users

Consider these scenarios:

Scenario 1: AI Medical Misdiagnosis

An AI diagnostic system analyzes a patient’s medical images and fails to detect cancer. The patient, relying on the AI diagnosis, doesn’t seek treatment. The cancer progresses, causing harm.

Who is liable?

• The AI developer who created the algorithm?
• The hospital that deployed the system?
• The data providers whose images trained the AI?
• The doctor who relied on the AI recommendation?
• The patient who didn’t seek a second opinion?

Legal Analysis:

Product Liability (Consumer Protection Act, 2019): The AI system might be a defective “product.” However, AI continuously learns and adapts-unlike static products. How do we assess whether an AI system is “defective”?

Professional Negligence: Did the doctor breach the duty of care by relying on AI? Medical standards of care are evolving; some AI systems exceed human accuracy. Courts will need to determine whether reasonable doctors should rely on AI recommendations.

Informed Consent: Did the patient consent to AI-assisted diagnosis? Was the patient informed about AI limitations? Lack of informed consent could constitute medical negligence.

Causation: Can we prove the AI misdiagnosis caused harm? Perhaps the cancer was undetectable even with perfect diagnosis, or the patient would have ignored correct diagnosis anyway. Establishing causation with AI involves complex counterfactuals.

My Current Legal Advice: Hospitals deploying medical AI should:

• Obtain explicit informed consent mentioning AI use
• Require human physician review of all AI recommendations
• Maintain detailed documentation of AI decisions and human oversight
• Ensure physicians are trained in AI limitations
• Obtain comprehensive liability insurance

However, this advice is conservative and defensive-reflecting legal uncertainty rather than clear legal standards.

Scenario 2: AI Credit Denial

A bank’s AI credit scoring system denies a loan application. The applicant suspects discrimination-the AI may have considered protected characteristics like caste, religion, or gender, even indirectly through proxy variables.

Legal Questions:

Discrimination: The Digital Personal Data Protection Act, 2023, requires fairness in automated decision-making. But does “fairness” prohibit disparate impact on protected groups? Indian law lacks clear algorithmic anti-discrimination standards.

Transparency: Does the applicant have a right to explanation of the AI decision? European GDPR provides “right to explanation” for automated decisions. Indian law is silent.

Accountability: Can the applicant challenge the decision? What evidence must the bank provide to justify the AI’s conclusion?

My Legal Advice:

Document fairness testing: Banks should conduct regular audits testing whether AI systems produce discriminatory outcomes across protected categories. Documentation demonstrates due diligence.

Provide explanations: Even without explicit legal requirements, provide applicants with meaningful explanations of AI decisions. This reduces litigation risk and builds trust.

Human oversight: Ensure high-stakes decisions (loan denials above certain thresholds) include human review, especially for borderline cases.

Challenge procedures: Establish clear procedures for applicants to challenge AI decisions and request human review.

Scenario 3: Autonomous Vehicle Accident

An autonomous vehicle causes an accident, injuring a pedestrian. The vehicle’s AI made a split-second decision to swerve, hitting the pedestrian while avoiding a larger collision.

Who is liable?

• The vehicle manufacturer?
• The AI software developer (potentially a different company)?
• The vehicle owner?
• The pedestrian who unexpectedly entered the road?

Legal Analysis:

Strict Liability: Indian tort law imposes strict liability for ultra-hazardous activities. Are autonomous vehicles ultra-hazardous? Courts will need to decide.

Product Defect: Was the AI’s decision evidence of a product defect? But perhaps no AI could have avoided all harm in that scenario. How do we assess whether an AI made an “unreasonable” decision in a no-win scenario?

Trolley Problem: The accident raises ethical questions (should AI prioritize passenger safety over pedestrian safety?) that become legal questions in negligence analysis.

Insurance: India’s Motor Vehicles Act requires insurance, but policies were drafted for human drivers. How do insurance frameworks apply to autonomous vehicles?

Current Law: India currently requires human drivers for all vehicles. Autonomous vehicle deployment will require legislative amendments to the Motor Vehicles Act-hopefully addressing liability clearly before autonomous vehicles operate widely.

The Need for Clear Liability Frameworks

These scenarios demonstrate that India’s current liability frameworks are inadequate for AI systems. We need legislation addressing:

Strict Liability for High-Risk AI: AI systems in critical domains (healthcare, autonomous vehicles, law enforcement) should face strict liability standards-liability without requiring proof of negligence.

Duty to Monitor: Entities deploying AI should have ongoing duties to monitor system performance, detect errors, and intervene when necessary.

Proportional Liability: Multiple parties share responsibility for AI harms. Liability should be allocated proportionally based on:

• Control over the AI system
• Benefit derived from the AI system
• Ability to detect and prevent harm
• Contribution to the harm

Mandatory Insurance: High-risk AI deployments should require insurance, ensuring victims can recover damages even if individual defendants lack resources.

Compensation Funds: For systemic AI harms affecting many people, India should consider compensation funds financed by AI industry contributions.

Until such frameworks exist, AI-related litigation will involve extensive legal uncertainty, inconsistent outcomes, and inadequate remedies for victims.

Data Privacy and AI: Navigating the Digital Personal Data Protection Act, 2023

The Digital Personal Data Protection Act (DPDP Act), 2023, represents India’s most significant data protection legislation. As a lawyer advising on AI deployments, I spend considerable time analyzing how the DPDP Act applies to AI systems.

Key Provisions Affecting AI

Consent Requirements (Section 6): Data fiduciaries must obtain consent before processing personal data. However, AI training often requires massive datasets-obtaining individual consent from millions of data subjects is impractical.

Purpose Limitation (Section 4): Personal data collected for one purpose generally cannot be used for another purpose without fresh consent. But AI training often involves repurposing data collected for other purposes (e.g., using customer service data to train chatbots).

Data Quality (Section 8): Data fiduciaries must ensure data processed for AI-driven decisions is complete, accurate, and consistent. This creates obligations to validate training data quality and address data errors.

Children’s Data (Section 9): Processing children’s data requires verifiable parental consent and prohibits behavioral tracking or targeted advertising. AI systems interacting with children (educational platforms, gaming) face heightened compliance burdens.

Data Principal Rights (Sections 13-14): Individuals have rights to:

• Access their personal data
• Correct inaccurate data
• Erase data (with exceptions)
• Nominate someone to exercise rights after death

Application to AI: Can individuals request deletion of their data from AI training datasets? This is technically challenging-once data trains a model, removing individual data points without retraining the entire model may be impossible.

Challenges in Applying DPDP Act to AI

Anonymization and AI: The DPDP Act applies only to “personal data”-data about identified or identifiable individuals. Anonymized data falls outside the Act’s scope.

However, AI can potentially re-identify anonymized data by correlating datasets. What constitutes sufficiently anonymized data in the AI era? The Act doesn’t specify, creating legal uncertainty.

Algorithmic Transparency: The Act requires data fiduciaries to provide information about data processing but doesn’t explicitly require explaining algorithmic decisions. Should AI deployers explain how algorithms reach conclusions? The Act is silent.

Cross-Border Data Transfers: The Act restricts transferring personal data outside India unless the destination country provides adequate data protection. However, AI systems often involve distributed processing across multiple jurisdictions. How do cross-border transfer rules apply? We await detailed regulations.

Trade Secrets vs. Transparency: Data principals have rights to information about data processing, but revealing detailed algorithmic logic may disclose trade secrets. How do we balance these competing interests? The Act doesn’t address this tension.

Practical Compliance Advice

Consent Mechanisms: Design consent mechanisms specifically for AI use cases:

• Be explicit about AI training purposes
• Provide granular consent options
• Implement consent management systems tracking consent status

Data Minimization: Collect only data necessary for AI purposes. Avoid retaining data indefinitely “just in case”-this increases compliance burdens and breach risks.

Privacy by Design: Integrate privacy protections into AI systems from inception:

• Use privacy-enhancing technologies (differential privacy, federated learning)
• Conduct Privacy Impact Assessments before deploying AI
• Implement technical measures limiting data processing to specified purposes

Data Security: Section 8 requires reasonable security safeguards. AI systems are attractive targets for cyberattacks. Implement:

• Encryption for data at rest and in transit
• Access controls limiting who can access training data
• Regular security audits and penetration testing
• Incident response plans for data breaches

Documentation: Maintain comprehensive records of:

• Data sources and collection methods
• Consent records
• Data processing purposes
• AI training methodologies
• Human oversight mechanisms
• Fairness and bias testing results

Accountability Mechanisms: Designate Data Protection Officers (if required by regulations); establish clear internal procedures for handling data principal rights requests; conduct regular compliance audits.

Sector-Specific Regulatory Challenges

Different sectors face unique AI-related legal challenges based on existing regulatory frameworks and the nature of AI deployments.

AI in Healthcare: Navigating Medical Device Regulations

Regulatory Gap: AI-powered diagnostic tools, treatment recommendation systems, and predictive analytics are increasingly common in Indian healthcare. However, India lacks clear regulatory frameworks specifically for AI medical devices.

Current Regulation: Medical devices are regulated under the Drugs and Cosmetics Act, 1940, and Medical Devices Rules, 2017. These regulations classify medical devices by risk and require approvals from the Central Drugs Standard Control Organization (CDSCO).

Challenges for AI Medical Devices:

Classification Uncertainty: How should AI diagnostic systems be classified? As medical devices? Software? The classification determines regulatory requirements.

Continuous Learning: Traditional medical devices are static-once approved, they don’t change. AI systems continuously learn and adapt. Do algorithm updates require fresh regulatory approval? Current regulations don’t address this.

Clinical Validation: Medical device approval requires demonstrating safety and efficacy through clinical trials. How do we validate AI systems that process vast datasets and produce probabilistic outputs rather than deterministic results?

Liability for AI Errors: When AI diagnostic systems produce incorrect diagnoses, who is liable? Current medical negligence law assumes human medical judgment. How do we assess whether AI-assisted medical decisions meet the standard of care?

My Legal Advice:

Proactive Engagement: Healthcare AI developers should engage with CDSCO early, seeking guidance on classification and approval pathways.

Treat AI as Medical Devices: Assume AI diagnostic or treatment systems are medical devices requiring regulatory approval. The consequences of non-compliance are severe.

Document Clinical Validation: Conduct rigorous clinical validation studies demonstrating AI accuracy, safety, and clinical utility. Peer-reviewed publications strengthen regulatory submissions.

Implement Human Oversight: Ensure AI recommendations are reviewed by qualified medical professionals before affecting patient care. This reduces liability risk and ensures compliance with medical standards.

Informed Consent: Obtain explicit patient consent for AI-assisted diagnosis or treatment, explaining AI’s role, limitations, and accuracy.

AI in Finance: RBI’s Evolving Framework

Current Approach: The Reserve Bank of India is addressing AI risks in banking and financial services through circulars, guidelines, and supervisory expectations rather than comprehensive regulations.

Key RBI Concerns:

Algorithmic Credit Scoring: AI credit scoring may perpetuate historical discrimination or create new forms of bias. RBI expects banks to ensure fairness, transparency, and explainability in AI-driven credit decisions.

Operational Resilience: AI systems create dependencies on technology vendors, data sources, and computational infrastructure. RBI expects robust business continuity and disaster recovery plans.

Model Risk Management: AI models can become inaccurate over time due to changing economic conditions, data drift, or unforeseen events. Banks must implement model validation, monitoring, and governance frameworks.

Customer Protection: AI-driven financial advice (robo-advisors) must be transparent, suitable for customer profiles, and subject to human oversight for complex or high-stakes decisions.

My Legal Advice:

Governance Frameworks: Establish board-level oversight of AI deployments. RBI expects financial institutions to have clear governance structures with senior management accountability.

Explainability: Ensure credit decisions can be explained to customers and regulators. Even if AI algorithms are complex, develop mechanisms to provide meaningful explanations.

Fairness Testing: Conduct regular audits testing whether AI systems produce discriminatory outcomes. Document testing methodologies and results for regulatory inspections.

Vendor Management: If using third-party AI solutions, ensure contracts clearly allocate responsibilities for model accuracy, data security, and regulatory compliance. Financial institutions remain accountable even when outsourcing AI.

Customer Disclosures: Inform customers when AI influences or makes financial decisions. Provide clear channels for customers to request human review.

AI in Law Enforcement: Balancing Security and Civil Liberties

Current Deployments: Indian law enforcement agencies increasingly use facial recognition systems, predictive policing algorithms, and AI-powered surveillance technologies.

Legal Concerns: These deployments raise profound civil liberties concerns, yet India lacks specific legal frameworks governing law enforcement AI.

Constitutional Issues:

Right to Privacy (Justice K.S. Puttaswamy v. Union of India, 2017): The Supreme Court established privacy as a fundamental right under Article 21. AI-powered surveillance potentially infringes privacy rights.

Equality (Articles 14-16): Biased AI systems in law enforcement could discriminate against marginalized communities, violating constitutional equality guarantees.

Due Process: Criminal justice decisions affecting liberty must follow due process. Can AI evidence be challenged? What standards govern AI’s role in investigations and prosecutions?

Lack of Legal Safeguards:

No Specific Legislation: Unlike facial recognition regulations in some EU countries, India lacks legislation governing law enforcement AI.

Minimal Judicial Oversight: Courts haven’t established clear standards for admissibility of AI-generated evidence, reliability requirements, or procedural safeguards.

Limited Transparency: Law enforcement AI deployments are often opaque-citizens don’t know what systems are used, how they work, or what data they process.

My Legal Perspective:

Urgent Legislative Need: India must enact legislation specifically governing law enforcement AI, establishing:

• Clear legal bases for AI deployments
• Robust oversight mechanisms (judicial, parliamentary, independent bodies)
• Transparency requirements
• Individual rights to challenge AI-generated evidence
• Accountability mechanisms when AI systems cause harm

Constitutional Litigation: Civil liberties organizations should consider strategic litigation challenging law enforcement AI deployments lacking legal authorization or adequate safeguards.

Admissibility Standards: Courts should establish clear evidentiary standards for AI-generated evidence-requiring validation of algorithms, testing for bias, and transparency about methodologies.

AI in Education: Protecting Student Data

Growing AI Use: Educational technology (ed-tech) platforms increasingly use AI for personalized learning, adaptive assessments, behavioral analysis, and administrative automation.

Legal Gap: India lacks specific regulations governing educational data privacy, algorithmic transparency in learning systems, or accountability when AI-powered assessments produce errors.

Concerns:

Sensitive Data: Ed-tech platforms collect sensitive student data-learning patterns, assessment scores, behavioral information, biometric data (facial recognition for proctoring). This data requires heightened protection.

Algorithmic Bias: AI-powered adaptive learning or proctoring systems may disadvantage certain student populations due to cultural biases in training data or algorithm design.

Long-Term Impacts: Educational AI decisions can affect students’ academic trajectories, career opportunities, and life outcomes. Errors can have profound consequences.

My Legal Recommendations:

Specific Legislation: India should enact educational data protection laws similar to the US Family Educational Rights and Privacy Act (FERPA), establishing:

• Restrictions on educational data collection and use
• Parental consent requirements
• Student rights to access and correct educational records
• Prohibitions on commercial use of student data

Algorithmic Audits: Ed-tech platforms should conduct regular algorithmic audits testing for bias across diverse student populations.

Transparency: Students and parents should understand how AI influences educational decisions, with clear explanations and opportunities to challenge AI-generated assessments.

Human Oversight: High-stakes educational decisions (exam scores, graduation determinations, university admissions) should involve human judgment, not solely AI.

The Deepfake Dilemma: Regulating Synthetic Media

Deepfakes-AI-generated synthetic media that convincingly impersonate real people-present some of the most urgent legal challenges I encounter.

The 2025 IT Rules Amendment

In 2025, MeitY amended the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021, specifically addressing synthetic content.

Key Provisions:

Mandatory Disclosure (Rule 3(1)(b)(v)): Social media users posting AI-generated synthetic content or modified content must disclose this fact, with at least 10% of visual display area or initial 10% of audio duration devoted to disclaimers.

Platform Responsibilities (Rule 3(1)(b)(vi)): Online platforms must:

• Implement clear labels identifying synthetic content
• Add warnings for suspected synthetic content
• Remove non-consensual intimate deepfakes within 24 hours of notice

Penalties: Circulating non-consensual deepfakes is punishable with up to 5 years imprisonment and significant fines under Section 66E of the IT Act.

Practical Challenges

Detection: How do platforms detect synthetic content? Current AI detection tools have high error rates-flagging authentic content as fake or missing sophisticated deepfakes.

User Compliance: How do platforms enforce mandatory disclosure requirements? Users may ignore or circumvent disclosure requirements.

Jurisdiction: Deepfakes often originate from anonymous sources or foreign jurisdictions. How do Indian authorities enforce penalties against offshore actors?

Defining “Synthetic”: What level of AI assistance triggers disclosure requirements? All AI-edited images? Only fully synthetic content? The rules lack precision.

Balancing Speech and Harm Prevention

Deepfake regulation involves delicate constitutional balancing:

Legitimate Uses: Not all deepfakes are harmful. Deepfakes are used for satire, political commentary, artistic expression, entertainment, and education-activities protected under Article 19(1)(a) (freedom of speech).

Harmful Uses: Deepfakes are used for non-consensual intimate imagery, fraud, election interference, defamation, and impersonation-activities causing serious harm.

My Legal Analysis:

Content-Neutral Regulation: Disclosure requirements are content-neutral-they don’t restrict what can be said but require transparency about how it’s said. This likely survives constitutional scrutiny.

Targeted Prohibitions: Banning non-consensual intimate deepfakes is a reasonable restriction under Article 19(2)-protecting dignity and preventing harm.

Over-Broad Restrictions: Blanket bans on all deepfakes would likely be unconstitutional-unduly restricting legitimate speech.

Procedural Safeguards: Takedown requirements must include procedural safeguards-notice, opportunity to respond, judicial review-to prevent censorship abuse.

Emerging Litigation

I anticipate significant deepfake litigation in coming years:

Defamation: Victims of deepfakes depicting them saying or doing things they didn’t will sue for defamation. Key question: Are deepfake creators liable even if they label content as synthetic?

Right of Publicity: Celebrities will sue for unauthorized use of their likeness in deepfakes. Indian law recognizes personality rights, but scope and remedies remain underdeveloped.

Electoral Offenses: Deepfakes manipulating election information may violate the Representation of the People Act, 1951. Election Commission will face challenges distinguishing legitimate political satire from illegal manipulation.

Criminal Prosecutions: Non-consensual intimate deepfakes will lead to criminal prosecutions under IT Act Section 66E and potentially rape laws (given evolving jurisprudence on digital sexual violence).

My Advice to Clients:

If Creating Deepfakes: Always label synthetic content clearly; obtain consent when using real people’s likenesses; avoid non-consensual intimate content absolutely; consider disclaimers even for obvious satire.

If Victimized by Deepfakes: Act quickly-document evidence; send takedown notices to platforms; file police complaints under IT Act Section 66E; consider civil defamation or personality rights claims; seek injunctions preventing further distribution.

Cross-Border Challenges: Jurisdiction and Enforcement

AI’s inherently global nature creates jurisdictional nightmares for lawyers and courts.

The Jurisdictional Question

The Problem: AI systems often involve:

• Developers based in one country (e.g., OpenAI in the USA)
• Servers and training infrastructure in another country
• Users and affected parties in India
• Data sources from multiple countries

When harm occurs, which country’s courts have jurisdiction? Which country’s laws apply?

Indian Courts’ Approach

Indian courts have generally adopted a long-arm jurisdiction approach for internet-related cases, asserting jurisdiction over foreign entities whose activities affect India.

Key Precedents:

Banyan Tree Holding v. A. Murali Krishna Reddy (2009): Delhi High Court held that courts have jurisdiction where “cause of action” arises-including where content is accessed, even if uploaded abroad.

MySpace Inc. v. Super Cassettes Industries Ltd. (2017): Delhi High Court exercised jurisdiction over MySpace (a US company) for copyright infringement, reasoning that MySpace’s services were accessible in India and affected Indian plaintiffs.

Application to AI: Following this jurisprudence, Indian courts will likely assert jurisdiction over foreign AI companies whose systems:

• Are accessible to Indian users
• Process data of Indian residents
• Cause harm to Indian individuals or entities
• Compete in Indian markets

This approach is consistent with ANI v. OpenAI, where the Delhi High Court is likely to reject OpenAI’s jurisdictional challenge despite OpenAI’s foreign location.

Enforcement Challenges

Jurisdictional Authority vs. Practical Enforcement: Even when Indian courts assert jurisdiction, enforcing judgments against foreign entities is challenging.

Asset Location: If defendants have no assets in India, enforcing monetary judgments requires foreign court recognition-which may not be granted if foreign courts disagree with Indian jurisdictional assertions.

Injunctive Relief: Courts can order blocking of websites or services within India, but foreign companies can circumvent blocks or may simply ignore orders.

Reciprocity: India is not party to comprehensive international enforcement treaties for civil judgments, making cross-border enforcement depend on bilateral arrangements or comity principles.

My Practical Advice:

For Indian Plaintiffs: Sue foreign AI companies in India if they have:

• Indian subsidiaries or offices (ensuring enforceability)
• Assets in India
• Contracts with Indian entities
• Regular business presence in India

If foreign defendants lack Indian presence, consider parallel litigation in their home jurisdictions or international arbitration if contractual provisions exist.

For Foreign AI Companies: If serving Indian markets:

• Establish Indian legal entities to clarify jurisdictional issues
• Include choice of law and arbitration clauses in terms of service
• Comply with Indian laws proactively to minimize litigation risk
• Maintain legal counsel familiar with Indian regulatory landscape

Data Localization and Sovereignty

Data Localization Requirements: Various Indian laws and regulatory proposals require certain data to be stored or processed within India:

Digital Personal Data Protection Act, 2023: Empowers the government to designate certain data categories requiring Indian storage. Detailed rules are pending.

RBI Payment Data Localization (2018): Requires payment system data to be stored exclusively in India.

Implications for AI: If AI systems process localized data, training and inference may need to occur within India-significantly affecting AI system architecture and costs.

Data Sovereignty: India, like many nations, increasingly asserts data sovereignty-the principle that data generated within a country is subject to that country’s laws and should benefit its citizens.

AI Context: This principle suggests:

• Training data originating in India should be used to benefit Indian AI development
• AI systems serving Indians should be accountable under Indian law
• Revenues from AI services to Indian users should be taxed in India

These principles will increasingly shape India’s AI regulatory approach, potentially creating compliance challenges for global AI companies but opportunities for domestic AI industry.

The Judicial Capacity Challenge: Can Courts Keep Pace?

As a litigator, I’m acutely aware that effective legal frameworks require not just good laws but competent judicial interpretation and enforcement. AI litigation presents unique challenges for Indian courts.

Limited Judicial Precedent

The Current State: Indian courts have decided very few cases directly addressing AI-specific legal issues. Beyond ANI v. OpenAI (still pending) and scattered cases touching algorithmic issues tangentially, we lack comprehensive AI jurisprudence.

Implications:

Doctrinal Uncertainty: Lower courts lack precedential guidance, leading to inconsistent decisions and prolonged litigation as parties appeal seeking authoritative interpretations.

Risk-Averse Litigation Strategy: Without clear precedents, risk-averse clients avoid litigation even when they have strong claims, preferring settlements to uncertain judicial outcomes.

Delayed Legal Development: Common law systems develop through case-by-case adjudication. Limited AI litigation means legal doctrines develop slowly-lagging behind technological reality.

Technical Complexity and Judicial Understanding

The Challenge: AI litigation requires judges to understand:

• Machine learning architectures (neural networks, deep learning, transformers)
• Training methodologies (supervised, unsupervised, reinforcement learning)
• Statistical concepts (probability distributions, overfitting, bias-variance tradeoffs)
• Computational processes (gradient descent, backpropagation)
• Specific AI applications (natural language processing, computer vision, recommender systems)

This level of technical sophistication exceeds what judges typically encounter in traditional legal disputes.

Current Reality: Indian judges-already burdened with massive caseloads-rarely have specialized technical training. While intellectually capable, they depend on expert witnesses, technical assessors, and party explanations to understand AI systems.

Problems This Creates:

Manipulation Risk: Parties with superior technical resources can present misleading technical explanations, exploiting judicial knowledge gaps.

Oversimplification: Judges may oversimplify complex technical issues, producing legally sound but technologically naive decisions that fail to address real problems or create unworkable standards.

Prolonged Proceedings: Technical disputes require extensive expert testimony, multiple hearings, and iterative questioning-extending already lengthy litigation timelines.

Inconsistent Outcomes: Different judges understand AI differently, leading to inconsistent decisions on similar facts.

International Models for Specialized Adjudication

Other jurisdictions have addressed this challenge through specialized mechanisms:

Specialized Courts: Some countries establish technology courts staffed by judges with technical training or expertise:

• Germany’s Patent Courts: Include technically trained judges for patent disputes
• UK’s Intellectual Property Enterprise Court: Handles IP cases with simplified procedures and technically informed judges

Expert Assessors: Courts appoint neutral technical experts who advise judges on complex technical matters:

• European Patent Office: Uses technical boards of appeal with scientific expertise
• Australian Federal Court: Frequently appoints expert assessors in technical cases

Technical Amicus Curiae: Organizations with technical expertise file amicus briefs explaining technical issues:

• US Supreme Court: Frequently receives technical amicus briefs from engineering organizations, tech companies, and academic institutions

My Recommendations for India

Establish Specialized Technology Benches: High Courts should designate specific benches to handle AI and technology cases, allowing judges to develop expertise through repeated exposure.

Technical Training Programs: Judicial academies should offer training programs on AI fundamentals, data science basics, and technology governance, enabling judges to engage meaningfully with technical evidence.

Court-Appointed Experts: In complex AI cases, courts should routinely appoint neutral technical experts-computer scientists, data scientists, AI ethicists-to provide independent assessments, reducing reliance on partisan expert witnesses.

Simplified Technical Explanations: Courts should require parties to submit plain-language explanations of technical issues alongside detailed technical evidence, ensuring judges grasp essential concepts.

Interdisciplinary Advisory Panels: The Supreme Court or Law Commission should establish standing advisory panels comprising legal scholars, technologists, ethicists, and industry representatives to provide guidance on emerging AI legal issues.

Ethical Challenges: The Lawyer’s Dilemma

Beyond analyzing law, practicing AI law confronts me with profound ethical dilemmas.

Advising on Legally Uncertain Ground

The Dilemma: Clients need definitive legal advice to make business decisions. Yet AI law is evolving rapidly-what’s legally acceptable today may be prohibited tomorrow; what seems prohibited today may be permitted after judicial interpretation.

The Tension: As lawyers, we’re trained to provide clear, confident guidance based on precedent and statutory interpretation. But with AI, precedents are sparse and statutes are ambiguous or nonexistent.

My Approach:

Transparency About Uncertainty: I explicitly acknowledge legal uncertainty rather than feigning false confidence. Clients deserve honest assessments, even if inconvenient.

Risk Stratification: I provide risk-stratified advice-identifying low-risk approaches (likely compliant even under restrictive interpretations), moderate-risk approaches (compliant under reasonable interpretations but potentially challenged), and high-risk approaches (likely non-compliant if challenged).

Scenario Planning: I help clients prepare for multiple regulatory scenarios, building flexibility into AI deployments so they can adapt as law evolves.

Conservative Default: When in doubt, I counsel compliance with the most restrictive plausible interpretation-reducing legal risk even if competitively disadvantageous.

Facilitating Harm vs. Enabling Innovation

The Dilemma: Some AI applications that are currently legal-or legally ambiguous-may cause social harm:

• Surveillance systems infringing privacy
• Algorithmic systems perpetuating discrimination
• AI-generated content spreading misinformation
• AI systems displacing workers without adequate social safety nets

As lawyers, we enable these deployments by structuring legally compliant implementations, drafting protective contracts, and defending against challenges.

The Ethical Question: Does legal permissibility equal ethical defensibility? Should lawyers refuse to facilitate legally permissible AI deployments we consider socially harmful?

The Counter-Argument: Lawyers aren’t moral arbiters. Our role is to help clients navigate law, not impose our ethical preferences. Clients, regulators, and society determine what’s acceptable-not individual lawyers.

My Personal Position: I balance zealous client representation with social responsibility:

Transparency: I inform clients about potential social harms of their AI deployments, even if legal. Many clients genuinely want to act responsibly and appreciate candid counsel.

Refusal in Extreme Cases: I reserve the right to decline representations involving AI applications I consider profoundly harmful-even if technically legal. For instance, I wouldn’t help deploy AI systems explicitly designed to discriminate or surveil vulnerable populations.

Advocacy for Better Law: I actively participate in policy consultations, draft model legislation proposals, and advocate for legal reforms that balance innovation with social protection.

Harm Mitigation: When clients proceed with legally ambiguous AI deployments, I insist on harm mitigation measures-fairness testing, transparency mechanisms, human oversight, accountability structures-that reduce risks even if not legally mandated.

The Access to Justice Challenge

The Reality: AI legal expertise is expensive. Sophisticated legal counsel on AI matters requires both legal and technical knowledge-commanding premium fees.

The Inequality: Well-resourced corporations can afford comprehensive AI legal advice; startups, nonprofits, and individuals often cannot. This creates asymmetric legal warfare-powerful entities deploy AI with full legal protection while those harmed by AI lack resources to challenge it.

Impact on Jurisprudence: Courts primarily hear cases brought by well-resourced parties, biasing legal development toward corporate interests rather than public interest concerns.

My Response:

Pro Bono Work: I dedicate time to pro bono representations-advising nonprofits, representing individuals harmed by AI systems, and supporting public interest litigation challenging problematic AI deployments.

Knowledge Sharing: I publish articles, give talks, and provide educational resources making AI legal knowledge accessible beyond paying clients.

Advocacy for Legal Aid: I advocate for government-funded legal aid specifically for AI-related disputes, ensuring individuals can meaningfully exercise their rights.

Supporting Public Interest Organizations: I support organizations like the Software Freedom Law Center, Internet Freedom Foundation, and similar entities providing free legal assistance on technology issues.

Recommendations: A Lawyer’s Wishlist for Legislative Reform

Based on years of practicing AI law, I offer these recommendations for Indian lawmakers:

1. Enact Comprehensive AI Legislation

Priority: A standalone AI Act addressing:

• Risk-based classification of AI systems
• Clear liability frameworks allocating responsibility for AI harms
• Transparency and explainability requirements for high-risk AI
• Algorithmic auditing mandates with independent verification
• Individual rights to challenge AI decisions
• Regulatory authorities with technical expertise and enforcement powers

Timeline: This legislation should be enacted by 2026 at the latest. The current regulatory vacuum creates excessive uncertainty, hindering both innovation and consumer protection.

2. Amend Copyright Law for AI

Specific Amendments Needed:

Text and Data Mining Exception: Add explicit provisions to Section 52 permitting text and data mining for AI training, with safeguards:

• Commercial use requires compensation mechanisms
• Rights-holders can opt out
• Attribution requirements
• Special protections for creative industries

AI Authorship Provisions: Clarify authorship of AI-generated works:

• Require substantial human creative contribution for copyright protection
• Deny copyright for purely autonomous AI outputs (placing them in public domain)
• Establish sui generis protection for AI-generated databases

Licensing Frameworks: Establish collective management organizations for AI training licenses, enabling rights-holders to be compensated fairly while allowing AI developers legal certainty.

3. Establish Clear Liability Standards

Strict Liability for High-Risk AI: AI systems in critical domains (healthcare, autonomous vehicles, criminal justice) should face strict liability-plaintiffs need only prove harm and causation, not negligence.

Proportional Liability: When multiple parties contribute to AI harms (developers, deployers, data providers), establish clear frameworks for allocating liability proportionally.

Mandatory Insurance: Require high-risk AI deployments to carry insurance, ensuring victims can recover damages.

Burden of Proof Shifts: In algorithmic discrimination cases, shift burden to defendants to prove their AI systems don’t discriminate once plaintiffs establish prima facie cases.

4. Create Specialized Judicial Mechanisms

Technology Courts: Establish specialized technology courts or tribunals in major cities, staffed by judges with technical training.

Expert Panels: Create standing expert panels in High Courts, available for appointment in complex AI cases.

Fast-Track Procedures: AI disputes often involve rapidly evolving technology. Establish fast-track procedures ensuring timely resolutions.

5. Strengthen Algorithmic Accountability

Mandatory Impact Assessments: Require algorithmic impact assessments before deploying high-risk AI, similar to environmental impact assessments.

Bias Testing Requirements: Mandate regular bias testing for AI systems affecting fundamental rights or significant opportunities.

Audit Rights: Grant regulators and affected parties rights to audit AI systems, with appropriate trade secret protections.

Transparency Registers: Establish public registers of high-risk AI deployments, enabling transparency about what systems operate in critical domains.

6. Protect Workers Through AI Transition

Reskilling Mandates: Require large employers deploying AI to invest in worker reskilling programs.

Advance Notice: Mandate advance notice before AI-driven layoffs, providing workers time to transition.

Social Safety Nets: Strengthen unemployment insurance, portable benefits, and transition support for workers displaced by AI.

Consultation Requirements: Require consultation with worker representatives before major AI deployments affecting employment.

7. Establish AI Safety Institute

Functions:

• Conduct independent research on AI risks and benefits
• Develop technical safety standards
• Provide evidence-based policy recommendations
• Certify high-risk AI systems meeting safety standards
• Investigate AI incidents and recommend preventive measures

Independence: The Institute must be independent from both industry and political pressures, with secure funding and statutory authority.

8. International Cooperation

Multilateral Engagement: India should actively participate in international AI governance initiatives:

• UNESCO AI Ethics Recommendations
• OECD AI Principles
• UN AI governance discussions
• Bilateral cooperation on AI regulation with EU, US, and other major jurisdictions

Mutual Recognition Agreements: Negotiate agreements recognizing AI certifications and compliance assessments across jurisdictions, reducing duplicative compliance burdens.

Data Governance Frameworks: Establish international frameworks for cross-border data flows supporting AI development while protecting privacy and security.

Impact on Legal Practice: How AI is Changing Lawyering Itself

Finally, I reflect on how AI is transforming my own profession-a transformation I experience daily.

AI Tools in Legal Practice

Legal Research: AI-powered legal research tools (LexisNexis, Westlaw, Indian legal AI platforms) dramatically accelerate case law research, statute interpretation, and precedent analysis.

Benefits: I can research comprehensively in hours what once took days. AI identifies relevant cases I might have missed through traditional keyword searches.

Concerns: Over-reliance on AI research risks missing nuances, misunderstanding holdings, or accepting AI-generated citations without verification (as several recent cases of fabricated citations demonstrate).

Document Review: AI-powered document review tools analyze contracts, identify relevant provisions, flag anomalies, and suggest revisions.

Benefits: Massive efficiency gains in contract review, due diligence, and discovery-reducing junior associate hours and costs.

Concerns: AI may miss contextual issues requiring legal judgment. Complete delegation to AI without human review creates malpractice risks.

Legal Drafting: Generative AI tools draft contracts, legal notices, pleadings, and other legal documents based on templates and instructions.

Benefits: Accelerates routine drafting, enabling lawyers to focus on strategy rather than formatting.

Concerns: AI-generated drafts require careful review-they may contain errors, inappropriate clauses, or fail to address client-specific issues. Lawyers remain responsible for all filings, regardless of AI assistance.

The Changing Value Proposition of Legal Services

Historical Model: Lawyers sold time-billing hourly for legal work regardless of efficiency. More hours meant more revenue.

AI Disruption: AI dramatically reduces time required for research, document review, and drafting. If lawyers continue billing hourly, AI makes legal services more efficient but less profitable.

The Shift: The profession is moving toward value-based billing-charging based on outcomes, expertise, and judgment rather than hours. AI handles routine tasks; lawyers provide strategic thinking, negotiation, judgment, and creative problem-solving.

My Adaptation: I increasingly position myself as strategic counselor rather than task executor. Clients value:

• Judgment: Applying legal principles to novel factual scenarios
• Strategy: Developing legal strategies achieving business objectives
• Advocacy: Persuading courts, regulators, and opposing parties
• Risk Assessment: Evaluating legal risks and recommending mitigation
• Relationship Management: Building trust and understanding client needs

These capabilities-requiring empathy, creativity, strategic thinking, and persuasion-remain distinctly human, at least for now.

Ethical Obligations in AI-Assisted Practice

Competence: Lawyers must be competent in the tools we use, including AI tools. This requires understanding AI capabilities, limitations, and potential errors.

Supervision: Lawyers remain responsible for all work product, even if AI-assisted. We must carefully review and validate AI outputs.

Confidentiality: Using AI tools requires sharing client information with AI providers. We must ensure AI tools comply with confidentiality obligations and data security requirements.

Disclosure: Should we disclose AI use to clients? Jurisdictions differ, but transparency seems prudent-informing clients about AI assistance while assuring human oversight.

Avoiding Fabrication: Recent cases have shown AI tools generating fake citations, cases, and legal authorities. Lawyers must verify all AI-generated legal authorities-failure to do so constitutes malpractice and may result in sanctions.

The Future of Legal Practice

Job Displacement Concerns: Will AI replace lawyers? In the near term, no. AI will replace certain lawyer tasks (routine research, basic drafting, document review) but not lawyers themselves.

Transformation, Not Elimination: Legal practice will transform-junior associate roles focused on routine tasks will decline; demand for senior lawyers providing judgment, strategy, and client relationship will persist.

Access to Justice Opportunities: AI could democratize legal access, enabling:

• Automated document generation for routine matters (wills, simple contracts)
• AI-powered legal information platforms helping individuals understand rights
• Virtual legal assistants providing preliminary guidance
• Reduced costs for routine legal services

However, this requires regulatory frameworks ensuring AI legal services are accurate, ethical, and don’t create new harms through erroneous advice.