United States

The Fourteenth Amendment of the US Constitution requires the government to provide due process and equal protection of laws to all persons.

Under the Equal Protection Clause, government use of AI may not unjustly discriminate against citizens, particularly in areas such as law enforcement. Predictive policing systems trained on historical police data may reflect bias against protected classes. But modern equal protection doctrine generally requires discriminatory intent, not merely disparate impact. See Washington v. Davis, 426 U.S. 229, 238–42 (1976). Accordingly, state AI uses that produce unfairly discriminatory effects may not violate the Equal Protection Clause absent discriminatory intent.

The Due Process Clause bars the government from depriving people of life, liberty, or property without due process of law. It is therefore relevant when agencies use AI to evaluate employees, applications for government benefits, and similar matters. For example, a school district’s use of AI to evaluate teachers was held to violate due process when terminated teachers lacked meaningful access to the algorithm or its reasoning to challenge the decision. See Houston Fed’n of Teachers Local 2415 v. Houston Indep. Sch. Dist. 251 F. Supp 3d 1168, 1180 (S.D. Tex. 2020).

In the United States, civil rights derive from the Constitution, statutes, and case law rather than from a separate body of designated human rights. Because the Constitution largely protects against government action rather than conferring affirmative entitlements, parties invoking “human rights” should identify the specific statutory or constitutional right at issue.

In order to obtain patent protection, an invention must be new, useful, and non-obvious. Moreover, abstract ideas are not eligible for patent protection.

Because AI techniques are generally well known, patent novelty requires more than using AI. See 35 United States Code (U.S.C.) section 101. To be patentable, the invention must improve technology or apply AI outputs in a specific real-world way. See Aon Re, Inc. v. Zesty.Ai, Inc., 791 F. Supp. 3d 531, 539 (D. Del. 2025) (holding that machine-learning classifiers analyzing aerial photographs of buildings created a concrete technological solution). Simply using AI to perform a task or adding AI to an existing process is not enough.

Courts have repeatedly found claims abstract where an invention merely applies machine learning to a new field without an inventive concept or technological improvement. See Recentive Analytics, Inc. v. Fox Corp., 134 F.4th 1205, 1215 (Fed. Cir. Ct. App. 2025). Likewise, describing AI and its benefits without explaining how to implement the invention does not support patentability. See Longitude Licensing Ltd. v. Google LLC, 2025 U.S. App. LEXIS 10353, *8-9 (Fed. Cir. Ct. App. 2025).

An invention created solely by AI is not patentable. See Thaler v. Vidal, 43 F.4th 1207, 1212 (Fed. Cir. 2022). In November 2025, the U.S Patent and Trademark Office rescinded prior guidance that had applied joint-inventorship factors to AI-assisted inventions. Under the revised guidance, inventorship is determined under traditional case law: AI cannot be an inventor because it is not a person, and patent protection for AI-assisted inventions is available only where a human conceived the invention.

Original works of authorship are protected under 17 U.S.C. section 102(a), but wholly AI-generated content is not. Where AI generates part of a work, the human-authored portion may be protected. In 2024, the Copyright Office granted protection to an artwork called, A Single Piece of American Cheese, which it had previously refused to register. In Part 2 of its Artificial Intelligence Report, the Copyright Office explained that AI outputs may be copyrightable where human contributions sufficiently shape the expression, including through perceptible human authorship, creative selection or arrangement, or creative modification of outputs (www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf.)

The Copyright Act provides under 17 U.S.C. section 107, that certain uses of copyrighted material are not considered infringement if they meet the statutory criteria for “fair use.” Courts weigh four factors in considering whether the defendant has demonstrated that its use of copyrighted material is a fair use:

• Purpose of the use. Educational use is more likely to be fair than commercial use.
• Nature of the copyrighted work. Factual works receive less protection than highly creative works.
• Amount of the work used. Using a small, non-essential part is more likely to be fair; copying the majority of the work is less likely to prevail.
• Effect on the potential market also weighs heavily. Uses that harm the market for the original count against fair use.

Model developers invoke fair use to defend training on copyrighted material, with results that often turn on whether the use is transformative. See, for example, Kadrey v. Meta Platforms, Inc., 788 F. Supp. 3d 1026, 1060 (N.D. Cal. 2025); Thomson Reuters Enter. Ctr. GmbH v. Ross Intel. Inc., 765 F. Supp. 3d 382, 401 (D. Del. 2025). In a May 2025 prepublication version of Part 3 of its AI Report, the Copyright Office stated that training on large, diverse datasets will often be transformative, but fair use still depends on deployment; models that generate outputs resembling their training data may be less transformative and more likely to cause market harm.

Copyright plaintiffs also often argue that training AI models on copyrighted works violates section 1202(b)(1) of the Digital Millenium Copyright Act (DMCA), which bars intentional removal or alteration of copyright management information (CMI) where the actor knows or has reason to know doing so will induce, enable, facilitate, or conceal infringement. CMI includes identifying information about a work, its owner, and its terms of use.

Although data preprocessing for large-scale training often strips peripheral metadata and CMI, courts have been reluctant to impose DMCA liability absent evidence that the AI system generates identical outputs, omitting that CMI.

Under the Uniform Trade Secret Act, trade secrets must be economically valuable, subject to reasonable efforts to maintain their confidentiality, and not generally known or readily ascertainable by proper means. See 18 U.S.C. section 1839(3).

Trade secret protection depends on secrecy, which can become harder to maintain as inventions are commercialized and exposed. If trade secrets are embedded in a model, carefully crafted prompts may elicit their disclosure. See OpenEvidence Inc. v. Pathway Medical, Inc. et al, 1:-25cv-1047 (D. Mass. 2025).

Authors Guild, Inc. v. Google, Inc., 804 F.3d 202, 225 (2d Cir. 2015). This case arose from Google’s Library Project, which scanned entire books but showed only snippets. The court held this was fair use, establishing that highly transformative digital uses may be fair even when entire works are copied. The case remains important as it established that transformative uses of copyrighted material for model training can be a fair use.

• It clarified that transformative digital uses can qualify as fair use, even when entire works are copied.
• It set a precedent for developers to use copyrighted material to train models for limited, transformative uses.

Kadrey v. Meta Platforms, Inc., 788 F. Supp. 3d 1026, 1060 (N.D. Cal. 2025). The court held that Meta’s training of large language models on plaintiffs’ books was sufficiently transformative to qualify as fair use.

Thomson Reuters Enter. Ctr. GmbH v. Ross Intel. Inc., 765 F. Supp. 3d 382, 401 (D. Del. 2025). The court found that Ross’s use of Westlaw-based headnotes to build a competing legal research tool was not fair use.

Raw Story Media, Inc. v. OpenAI, 1:24-cv-01514 (S.D.N.Y. 2025). The court held the plaintiff lacked standing because it did not show that ChatGPT had distributed, or was likely to distribute, its copyrighted works without CMI.

Doe v. GitHub, Inc., 672 F. Supp. 3d 837, 851-52 (N.D. Cal. 2023). Plaintiffs alleged GitHub trained on public repositories and generated code stripped of notices, license terms, and attribution. The court dismissed the DMCA claims for lack of standing because plaintiffs had not shown the models reproduced identical licensed code. Whether identicality is required under section 1202(b)(1) is on interlocutory appeal to the Ninth Circuit.

The United States lacks a comprehensive federal data law, but numerous federal and state laws regulate the personal data used in AI systems, including:

• Federal Trade Commission (FTC) Act, which prohibits deceptive trade practices (such as breaching promises to consumers in privacy policies) and unfair practices (such as establishing inadequate security on personal data that was promised to be kept confidential, and using facial recognition technology in retail stores without taking reasonable steps to address the risk that the technology was likely to result in substantial injury to consumers that consumers cannot reasonably avoid themselves and that is not outweighed by countervailing benefits to consumers or competition as a result of false-positive facial recognition match alerts);
• California Consumer Privacy Act (CCPA), which governs the use of automated decision-making technology (ADMT) for “significant decisions” affecting consumers. Significant decisions include those affecting finances, housing, education, employment, or healthcare;
• Health Insurance Portability and Accountability Act (HIPAA), which limits sharing of sensitive health care patient data;
• Fair Credit Reporting Act (FCRA), which limits usage of data collected by consumer reporting agencies; and
• Children’s Online Privacy Protection Rule (COPPA), which prohibits collection or use of personal information from and about young children.

Furthermore, several US states have enacted comprehensive consumer data privacy laws that require businesses to obtain explicit consent before using certain categories of personal information. These laws generally follow an opt‑out model for most data uses but require opt‑in (consent) for sensitive or restricted categories. California has the strongest enforcement agency and private right of action. Other states rely on state attorneys general or FTC actions for enforcement.

The EU General Data Protection Regulation (GDPR) can apply to US companies that target the EU or process EU personal data. Companies transferring data to or from the EU should use the EU’s Standard Contractual Clauses for lawful transfers.

US companies operating in the EU should also be cautious about using publicly available EU personal data to train models. In March 2026, Italy’s data protection authority fined OpenAI GBP 9 million for using personal data to train ChatGPT without an adequate legal basis.

In 2024, the National Science Foundation launched the National AI Research Resource (NAIRR) pilot to provide access to software, data, models, and educational materials that support AI research and innovation. Following the White House AI Action Plan, NAIRR is establishing an Operations Center to transition the pilot into a sustainable long-term program promoting open and competitive US AI innovation.

The United States does not have a comprehensive federal law regulating biometric data. However, many states have enacted their own laws including:

• California Privacy Rights Act (CPRA), which creates notice and consent requirements before use and retention of biometric data and enables consumers to request businesses limit use of biometric data and deletion of biometric data;
• Colorado Privacy Act (CPA), which requires collectors to obtain consent before disclosing, redisclosing, or otherwise disseminating any consumer’s and employees biometric identifiers and creates restrictions on the sale, purchase and retention of biometric identifiers and data;
• Illinois’ Biometric Information Privacy Act (BIPA), which requires entities that collect biometric data to maintain a written policy about the collection, storage, and retention of biometric data and to obtain informed consent from individuals subject to biometric data collection;
• Texas’ Capture or Use of Biometric Identifier (CUBI), which requires advance notice prior to the collection of a biometric identifier for a “commercial purpose,” and consent to the collection and specific uses of the biometric identifier; and
• Virgina Consumer Data Protection Act (VCDPA), which requires controllers to obtain consent prior to processing consumer’s biometric data.

As noted in Section 1.1, above, state use of AI must comply with the Due Process Clause, and the Equal Protection Clause bars unlawful discrimination by the government, including through AI. Government decisions that incorporate AI must therefore satisfy constitutional anti-discrimination requirements.

The United States civil rights laws also extend anti-discrimination prohibitions to private AI use. These laws include:

• Equal Credit Opportunity Act;
• Title VII of the Civil Rights Act of 1964;
• Fair Credit Reporting Act;
• Federal Trade Commission (FTC) Act;
• Americans with Disabilities Act;
• Age Discrimination in Employment Act;
• Fair Housing Act; and
• Genetic Information Non-discrimination Act.

These civil rights laws apply whether discrimination is intentional or inadvertent. In FTC v. Rite Aid, the FTC alleged that Rite Aid made “no effort” to assess the accuracy or bias of its facial-recognition system in violation of section 5 of the FTC Act. When AI systems implicate civil rights laws, rigorous auditing and bias review are essential.

The Colorado Artificial Intelligence Act (CAIA) (effective June 30, 2026) regulates “high-risk” AI systems — those that make, or are a substantial factor in making, consequential decisions in areas such as employment, education, housing, health care, insurance, legal services, financial services, and essential government services.

The CAIA defines algorithmic discrimination broadly to include intentional discrimination and disparate impact based on protected characteristics, Disparate impact applies where  unlawful differential treatment or impact disfavors individuals or groups based on protected characteristics such as age, color, disability, ethnicity, genetic information, limited English proficiency, national origin, race, religion, reproductive health, sex, veteran status, or other protected traits under Colorado or federal law.

As a result, liability under the CAIA may arise even without intent to discriminate.

The CAIA has also drawn federal attention. On December 11, 2025, President Trump directed the Attorney General to establish an AI Litigation Task Force to identify and challenge state AI laws viewed as inconsistent with a “minimally burdensome national policy framework for AI.” See Exec. Order No. 14365, 90 C.F.R. 58499 (2025). President Trump cited Colorado’s law as part of the problem, stating, “the new Colorado law banning ‘algorithmic discrimination’ may even force AI models to produce false results in order to avoid a ‘differential treatment or impact’ on protected groups.”

The National Institute of Standards and Technology (NIST) administers a range of standards relevant to federal agencies and contractors. In December 2025, it released a preliminary draft Cybersecurity Framework Profile for Artificial Intelligence (NIST IR 8596), a voluntary profile that complements the NIST AI Risk Management Framework and Cybersecurity Framework 2.0. It focuses on securing AI components, using AI for cyberdefense, and countering AI-enabled attacks. Although voluntary, it offers best practices that may become a regulatory benchmark.

In June 2025, FTC Commissioner Melissa Holyoak noted that AI can both combat fraud and concentrate market power. The FTC is concerned that unequal access to training data, compute, and talent could concentrate the market among a small number of providers and suppress competition. At the same time, the White House AI Action Plan directs the FTC to ensure that investigations and orders do not unduly burden AI innovation, reflecting a policy balance between competition enforcement and national AI development.

Another concern is AI-enabled pricing collusion. Traditional price-fixing is addressed under section 1 of the Sherman Act and section 5 of the FTC Act, but regulators increasingly worry about tacit collusion, where firms use algorithms to monitor and respond to competitors’ pricing without any provable agreement. Because section 1 requires proof of agreement, such conduct may evade existing antitrust rules.

US antitrust law is governed chiefly by the Sherman Act, Clayton Act, and FTC Act. At the state level, California, New York, and Connecticut have enacted laws targeting algorithmic pricing. California bans use or distribution of a common pricing algorithm as part of a cartel or conspiracy, though it does not appear to reach competitors using separate tools. New York’s law focuses on residential real estate and bars algorithmic coordination of rental terms, while Connecticut prohibits use of revenue-management tools to set rents or occupancy levels in residential housing. These laws remain limited but reflect a broader trend toward state regulation in the absence of federal action.

Agentic AI refers to systems capable of autonomous action, interacting with their environment, and completing multi-step tasks. In February 2026, NIST announced an AI Agent Standards Initiative to develop voluntary, industry-led guidance for AI agents, signaling continued US investment in agentic AI.

US law has not yet defined the legal status of AI agents. However, the FTC’s August 2025 complaint against Air AI — alleging overstated claims about its agentic AI — shows that regulators are already scrutinizing agentic AI marketing and capabilities.

There are not currently any sector-specific agentic AI regulations or requirements for Agentic AI under US law.

It has not yet been established in the US whether AI is a product or a service from a products liability perspective. Still, the EU Product Liability Directive and the AI LEAD Act, introduced in the Senate in 2025, treat AI systems as products, suggesting policy interest in that approach. Even so, product-liability claims generally require physical harm, which may be difficult to prove for AI systems that are not autonomous vehicles/robots or a smart device.

AI may fall within strict liability when integrated into products, such as autonomous vehicles or smart devices. A product may be defective because of a manufacturing, design, or warning defect — for example, if an AI-enabled product deviates from design, is unreasonably dangerous because of flaws, such as unmitigated bias or hallucinations, or lacks adequate warnings about foreseeable risks. Courts will still need to determine which party is responsible for such defects.

Because US law does not yet clearly treat AI as a product for product-liability purposes, there are no settled AI-specific recall or safety requirements. However, the FTC Act still bars unfair or deceptive practices, including failure to address AI defects, and general case law imposes duties to warn of or remedy defects.

In 2025 the following cases involved AI and products liability theories:

• Garcia v. Character Techs., Inc., where the court treated the AI companion chatbot like a product under strict-liability pleadings following a 14-year-old’s suicide after forming an intense emotional relationship with the chatbot. See 785 F. Supp. 3d 1157, 1180 (M.D. Fla 2025).
• Benavides v. Tesla, Inc., where the court held that the plaintiffs provided enough evidence to establish that Tesla’s Autopilot system was a substantial factor in their injuries under a products liability theory. See 804 F. Supp. 3d 1242, 1299 (S.D. Fla 2025).
• Nevada v. MediaLab AI, Inc., in which the Nevada attorney general filed suit against the tech holding company of the app Kik for violating the Nevada Deceptive Trade Practices Act and brought claims of negligence, products liability and unjust enrichment. See Case No. A-25-92-5766-C (Nev. Dist. Ct. Clark County, 2025).

In absence of comprehensive federal legislation regulating AI, most AI legislation is governed by the individual states. In 2025, over 1,000 AI bills were introduced with 38 states enacting approximately 100 AI laws regulating AI development and use. For a substantially complete review of the pending and enacted AI legislation, see the National Conference of State Legislatures at www.ncsl.org/technology-and-communication/artificial-intelligence-2025-legislation. Of these enacted laws, a few notable state AI laws include:

• California’s Companion Chatbot law and Transparency in Frontier Artificial Intelligence Act (TFAIA);
• California Consumer Privacy Act (CCPA) and regulations governing automated decision-making;
• Colorado AI Act;
• New York’s Responsible AI Safety and Education Act (RAISE) Act;
• Texas Responsible Artificial Intelligence Governance Act (TRAIGA); and
• Utah’s AI Policy Act.

On January 20, 2025, President Trump revoked Biden’s Executive Order 14110 on AI safety. See Exec. Order No. 14179, 90 C.F.R. 8741 (2025). In July of 2025, the White House released America’s AI Action Plan available at www.whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-Action-Plan.pdf. The AI Action Plan accompanied three executive orders focusing on accelerating AI innovation, building AI infrastructure, and enhancing international competition. The plan creates funding opportunities and promotes private sector innovation.

To accelerate innovation, the plan directs the Federal Communications Commission to preempt state AI laws that burden AI innovation. Under the AI Action Plan, the FTC is instructed to review, modify, and set aside any investigations or orders that “unduly burden AI innovation.” Similarly, on December 11, 2025, President Trump signed a new executive order directing the Attorney General to establish an AI Litigation Task Force to identify and challenge state AI laws that are inconsistent with the “minimally burdensome national policy framework for AI.” See Exec. Order No. 14365, 90 C.F.R. 58499 (2025). The AI Action Plan encourages agencies to invest in developing and scaling foundational and translational manufacturing technologies and states that open-source models are critical for innovation, research, and competitiveness. The plan also creates procurement restrictions, procurement regulations and policies, and directs the NIST to focus on addressing the risks of deepfakes.

To streamline and incentivize building AI infrastructure on federal lands, Executive Order No. 14318 directs the Department of Interior, Department of Energy and Department of Commerce to authorize qualifying projects. See Exec. Order No. 14138, 90 C.F.R. 14212 (2025). Similarly, the Executive Order seeks to expedite the permitting process for data centers, energy generation and electricity transmission, and additional infrastructure to serve those projects. The AI Action Plan also requires agencies to update existing cybersecurity approaches and measures to account for AI. The AI Action Plan also directs federal agencies to support the development and deployment of American “AI Stack” globally.