Liability and Torts

PRODUCT LIABILITY AND TORT ISSUES INHERENT TO AI MODELS

This area of the law addresses the internal legal risks stemming directly from the design, function, and decision-making processes of a modern AI system, independent of its interaction with legacy data. These nuanced issues revolve around accountability, liability, and the challenge of auditing opaque algorithms.

Product Liability and Tort Law

The dominant litigation risk here is Product Liability and Tort Law. Modern AI systems, such as autonomous vehicles or medical diagnostic tools, are no longer static software but dynamic, learning "products." The legal question becomes: When an AI causes harm (e.g., an autonomous car collision, an incorrect medical diagnosis), does the established framework of strict liability apply? Strict liability holds a manufacturer responsible for defects regardless of fault. 

The defense often points to the AI's "learning" capability, arguing that a decision made autonomously by a self-improving system represents an unforeseeable intervening event, absolving the original developer. Litigators counter by focusing on design defect or failure to warn, arguing the developers were negligent in the training data selection, hyper-parameter tuning, or failure to impose adequate guardrails.
 

Algorithmic Discrimination and Bias

Closely related is the risk of Algorithmic Discrimination and Bias. Litigation in this space often invokes established anti-discrimination statutes, such as Title VII (employment) and the Equal Credit Opportunity Act (lending). The AI system, reflecting biases inherent in its historical training data, may generate discriminatory outcomes. 

Legally, proving disparate impact (where a neutral policy disproportionately harms a protected group) is easier than proving disparate treatment (intentional discrimination). The legal defense is often that the system is merely reflecting historical data without malicious intent. However, ethical and legal standards are converging on a duty to mitigate bias. 

Litigation focuses heavily on the developer's failure to conduct rigorous bias audits, failure to implement debiasing techniques, and a lack of transparency regarding the features and data used for decision-making.
 

Transparency, Explainability (XAI), and Due Process 

The inherent "black box" nature of many deep learning models creates a fundamental Transparency, Explainability (XAI), and Due Process challenge. When a critical AI system, used in areas like criminal sentencing, social services eligibility, or consumer credit scoring, makes a decision, the affected party has a legal right, and often a constitutional due process right, to understand the rationale. 

The technical difficulty in extracting a human-readable, causally accurate explanation from a high-dimensional model becomes a legal liability. Regulatory frameworks like the EU AI Act1 and the GDPR2 explicitly mandate rights of explanation and auditability, transforming a technical limitation into a non-compliance litigation risk. 

In litigation, a lack of transparency can lead to an adverse inference or even summary judgment against the deploying entity, as they may be unable to produce evidence demonstrating the system's compliance or lack of defect. This forces companies to invest heavily in auditable AI structures, including model cards and immutable logging of decision paths, anticipating future legal scrutiny.

DISCUSSION and CASE LAW

PRODUCT LIABILITY and TORT LAW

This area focuses on applying strict liability, design defects, and failure to warn to AI systems causing harm, such as in autonomous vehicles or medical diagnostics, with defenses emphasizing the AI's autonomous learning as an unforeseeable event. Relevant cases include:

• Taylor v. Intuitive Surgical, Inc. (Wash. Sup. Ct.      2017, holding that inadequate warnings for robotic surgical systems can      render a product defective if risks could be mitigated by reasonable      instructions to overseers)
• Terhune v. A.H. Robins Co. (Wash. Sup. Ct. 1978,      establishing that warnings in medical contexts should target physicians or      overseers due to manufacturers' challenges in reaching end users)
• McCourt v. J.C. Penney Co., Inc. (Nev. Sup. Ct. 1987,      incorporating alternative designs into defect assessments for consumer      expectations)
• Mazile v. Larkin Univ. Corp. (S.D. Fla. 2024,      dismissing discrimination claims against an AI testing system for lack of      evidence of known bias, but compelling arbitration for defect-related      issues)
• Hicks v. Collier (S.D. Tex. 2024, preserving claims      against AI housing classification for lack of transparency and accuracy,      emphasizing accountability in deployment).

ALGORITHMIC DISCRIMINATION AND BIAS

Litigation often relies on disparate impact under statutes like Title VII and the ECOA, targeting biases in training data without proving intent, with defenses claiming reflection of historical realities. Key cases include:

• Mobley v. Workday, Inc. (N.D. Cal. 2023-2025,      ongoing; denying dismissal and certifying collective action under ADEA for      AI screening tools causing age, race, and disability bias via disparate      impact)
• Saas v. Major, Lindsey & Africa, LLC (D. Md.      2024, dismissing Title VII and ADEA claims for insufficient pleading of AI      tool use in discriminatory recruitment)
• Harper v. Sirius XM (E.D. Mich. ongoing 2025,      alleging Title VII violations from AI hiring software using biased proxies      like employment history to disadvantage Black applicants)
• Equal Rights Center v. Meta (D.C. Super. Ct. 2025,      denying dismissal for disparate treatment and impact under D.C. Human      Rights Act from algorithmic ad delivery discrimination)
• Inclusive Communities Project v. Texas Department of      Housing & Community Affairs (U.S. Sup. Ct. 2015, affirming disparate      impact claims under Fair Housing Act for neutral policies causing      disparities).

TRANSPARENCY, EXPLAINABILITY (XAI), AND DUE PROCESS

Challenges arise from black-box models in high-stakes decisions, invoking due process rights and mandates under the GDPR and EU AI Act for explanations and auditability, with litigation risking adverse inferences for non-transparency. Pivotal cases encompass:

• State v. Loomis (Wis. Sup. Ct. 2016, addressing due      process failures in opaque AI risk assessments for sentencing)
• Williams v. City of Detroit (ongoing; highlighting      explainability issues in faulty facial recognition leading to false      arrests)
• Schufa (CJEU C-634/21, 2023, ruling credit scoring as      automated decision-making under GDPR Article 22 with legal effects if      relied upon heavily)
• D&B (CJEU C-203/22, 2025, confirming GDPR right      to meaningful explanations of automated logic, balancing with trade      secrets)
• Nacionalinis (CJEU C-683/21, 2023,      establishing joint controllership in automated processing)
• KNLTB (CJEU C-621/22, 2024, allowing legitimate      interests for automated processing if balanced)
• Ligue des droits humains (CJEU C-817/19, 2022,      limiting opaque AI in mass data analysis for lack of transparency and      remedy).

NOTE: CJEU is the Court of Justice of the European Union; SCHUFA in this instance refers to SCHUFA Holding AG, Germany credit bureau; D&B in this instance refers to Dun & Bradstreet Austria GmbH; Nacionalinis refers to: "National Public Health Centre under the Ministry of Health (Lithuania)"; KNLTB refers to Koninklijke Nederlandse Lawn Tennis Bond, the Royal Dutch Lawn Tennis Association; Ligue des droits humains("LDH") is the League of Human Rights," Belgian

FOOTNOTES

Footnote 1: The EU AI Act establishes a risk-based regulatory framework for artificial intelligence systems, classifying them as prohibited, high-risk, limited-risk, or minimal-risk to ensure safety, transparency, and protection of fundamental rights. It imposes obligations on providers and deployers, such as conformity assessments for high-risk AI, while promoting innovation and the free movement of AI technologies across the EU. Reference: Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024

Footnote 2: The General Data Protection Regulation (GDPR) is an EU law that establishes stringent rules for protecting personal data of individuals within the European Economic Area, emphasizing principles like transparency, accountability, and data minimization. It grants data subjects rights such as access, rectification, and erasure while imposing obligations on organizations, including data breach notifications and potential fines up to 4% of global annual turnover for non-compliance. Reference: Regulation (EU) 2016/679, available at https://eur-lex.europa.eu/eli/reg/2016/679/oj.