Judges Grapple with Algorithms That Test the First Amendment

Free speech law has never faced a defendant like this. As Congress drafts new rules for artificial intelligence, courts are being asked to decide whether a machine’s output deserves the same protection as a journalist’s article or a lawyer’s brief. For law firms and regulators, the implications reach beyond philosophy. If AI-generated work product counts as “speech,” then nearly every compliance safeguard, from disclosure to audit to preapproval, may face constitutional challenge.

From Encryption to AI Code as Protected Speech

For more than two decades, courts have extended First Amendment protection to software code and search engine rankings. In Bernstein v. U.S. Department of Justice (1999), the Ninth Circuit held that encryption source code could constitute protected speech. Later, in Search King v. Google Technology (2003), a federal court found that Google’s PageRank algorithm was a form of protected opinion. Together, those cases suggested that expression can exist without a human voice, a principle now tested by generative AI.

Generative AI extends that doctrine to its limit. When a model drafts contract language or document summaries, developers might claim First Amendment protection over the output. Legislators seeking to regulate AI disclosures risk being accused of prior restraint. For firms building or using legal AI tools, this is not theoretical. It defines what can be mandated, monitored, or audited without triggering a constitutional fight.

Copyright Litigation and Regulatory Firewalls

Current litigation illustrates the collision between regulation and expression. In P.M. v. OpenAI, filed in June 2023, plaintiffs allege that large language models unlawfully used personal information to generate content. OpenAI argues that training and output are expressive acts protected under the First Amendment and fair use doctrine. A similar claim appears in Doe v. GitHub, filed in November 2022, where developers assert that Copilot reproduces code from public repositories without attribution. The defense again rests on expressive freedom, arguing that if AI models speak through code, their training may be constitutionally shielded.

In December 2023, The New York Times sued OpenAI and Microsoft in federal court, alleging copyright infringement through unauthorized use of millions of articles to train AI models. The case, consolidated with suits from other publishers, proceeds to discovery after the court denied most motions to dismiss in April 2025. This litigation will determine whether training AI on copyrighted content constitutes fair use or infringement on an unprecedented scale.

The Department of Justice and several amici, including the Knight First Amendment Institute, have acknowledged the tension. Restricting model training or output could be viewed as content-based regulation. Yet the absence of rules leaves consumers, lawyers, and regulators without clear accountability. The AI Disclosure Act of 2023 (H.R. 3831) and the Generative AI Copyright Disclosure Act of 2024 (H.R. 7913) aim to require identification of AI-generated content and disclosure of copyrighted training data. Each faces constitutional objections that mandatory labeling compels speech.

For law firms, the effect is practical. If the First Amendment shields AI developers from compelled disclosure, internal compliance programs may need to treat model explanations as proprietary information. Vendor contracts, audit clauses, and disclosure requirements must therefore balance transparency with constitutional limits, a calculus rarely encountered in standard due diligence workflows.

Due Process and the Right to Explainable AI

While the First Amendment guards expression, the Fifth guarantees due process. AI systems used in sentencing, credit scoring, or hiring increasingly determine rights and obligations, often without transparent reasoning. The Blueprint for an AI Bill of Rights (White House OSTP, 2022) warns that automated systems must remain explainable and contestable. That principle is rapidly evolving from policy to constitutional expectation.

Courts first addressed this concern in State v. Loomis (2016), where the Wisconsin Supreme Court upheld use of a proprietary risk assessment at sentencing but warned that opacity could violate due process. Federal agencies now apply similar reasoning. The Department of Justice’s December 2024 AI and Criminal Justice Final Report urges that AI tools enhance, not replace, human judgment. The Federal Trade Commission likewise warns that unexplainable models may constitute deceptive practices under existing law.

For firms deploying AI in research or client services, the Fifth Amendment’s principle of reasoned decision-making has become a governance mandate. Documentation of model provenance, accuracy testing, and human review are now evidence of procedural fairness. When clients challenge an adverse outcome tied to AI, due process obligations can overlap with professional responsibility and malpractice risk.

Law firms building internal tools or advising clients must treat explainability as both a technical and constitutional safeguard. Without verifiable reasoning, firms risk claims of arbitrary or discriminatory practice under procedural rights doctrines. What began as consumer protection is evolving into a constitutional compliance framework.

Compelled Labeling as Prior Restraint

Congressional efforts to regulate AI transparency now face constitutional headwinds. Compelled disclosure requirements risk being framed as restrictions on expression. When the AI Disclosure Act proposed that generative AI output include visible identifiers, free speech advocates argued it forced private actors to label their expression. Industry groups cited Virginia State Board of Pharmacy v. Virginia Citizens Consumer Council (1976), which struck down restrictions on truthful commercial speech. In this new context, mandatory labels could face similar scrutiny.

Regulators counter that disclosure is essential for accountability. Without transparency, the AI marketplace, including tools used in legal practice, operates without informed consent. The resulting standoff is visible in hearings where legislators call for risk mitigation while constitutional scholars warn of chilled innovation. The absence of a federal standard ensures every AI governance rule carries litigation potential from day one.

For law firms, this tension creates operational challenges. If disclosure mandates are constitutionally limited, firms cannot rely solely on regulation to verify vendor integrity. They must design private equivalents: internal labeling, audit rights, and usage logs that satisfy client and insurer demands. Constitutional uncertainty thus migrates into contract law and compliance design.

AI Immunity and Liability

When Congress passed the Communications Decency Act in 1996, it could not foresee algorithms that write, reason, and decide. Section 230 shielded online platforms from liability for user-generated content. Generative AI raises a harder question: is the model a user, a publisher, or something in between? The answer will determine whether existing immunity doctrines survive.

Policy analysts at the Stanford Cyber Policy Center and the Brookings Institution note that future legislation could introduce a conditional shield tied to documented safety and provenance standards. That approach would link constitutional protection to governance behavior: firms maintaining audit trails and human oversight could retain immunity, while opaque systems would not.

This framework would merge constitutional rights with technical governance. Free speech would remain protected, but only for systems designed to prove accountability. For legal practitioners, this convergence sets a new professional baseline. Speech is defensible only when traceable. Law firms integrating AI into workflows will need model registers, version logs, and bias assessments to show that speech was both free and responsible.

Malpractice Risk and AI Governance Standards

As AI regulation accelerates, firms now operate where constitutional rights and compliance duties intersect. Generative platforms that draft filings or client communications may be constitutionally expressive, yet they expose lawyers to malpractice and discovery risk. The ABA Formal Opinion 512, issued July 29, 2024, requires reasonable oversight of AI-assisted work, emphasizing human review and recordkeeping. But if developers claim constitutional protection over their models, attorneys may face limits in auditing or disclosing how outputs were produced.

Insurance carriers are responding. Cyber liability underwriters now request documentation of AI risk controls, human review policies, and vendor due diligence. Verified governance frameworks such as the NIST AI Risk Management Framework and ISO/IEC 42001 certification can lower premiums. The result is a dual compliance economy: one driven by regulation and the other by insurance incentives. In both, constitutional ambiguity influences the cost of assurance.

For Washington, the friction between free speech and fair process is no longer theoretical. It defines how far agencies can push oversight without suppressing innovation, and how much responsibility private actors must bear when AI becomes their voice. For law firms, it marks a turning point. The First Amendment is no longer just a shield for journalists. It is a governance challenge for every professional who deploys a model that speaks.

Sources

• American Bar Association: Formal Opinion 512 (July 29, 2024)
• Bernstein v. U.S. Department of Justice, 176 F.3d 1132 (9th Cir. 1999)
• Brookings Institution: How Different States Are Approaching AI
• Blueprint for an AI Bill of Rights (White House OSTP, 2022)
• Congress.gov: AI Disclosure Act of 2023 (H.R. 3831)
• Congress.gov: Generative AI Copyright Disclosure Act of 2024 (H.R. 7913)
• Doe v. GitHub, Inc., 4:22-cv-06823 (N.D. Cal. 2022)
• Federal Trade Commission: Misleading AI Claims (2024)
• ISO/IEC 42001: Artificial Intelligence Management System Standard (2023)
• Knight First Amendment Institute at Columbia University
• NIST AI Risk Management Framework (2023)
• P.M. v. OpenAI LP, 3:23-cv-03199 (N.D. Cal. 2023)
• Search King v. Google Technology, Inc., 2003 WL 21464568 (W.D. Okla. 2003)
• Stanford Cyber Policy Center
• State v. Loomis, 881 N.W.2d 749 (Wis. 2016)
• The New York Times Company v. Microsoft Corporation, 1:23-cv-11195 (S.D.N.Y. 2023)
• U.S. Department of Justice: Artificial Intelligence and Criminal Justice Final Report (December 2024)
• Virginia State Board of Pharmacy v. Virginia Citizens Consumer Council, 425 U.S. 748 (1976)

This article was prepared for educational and informational purposes only. It does not constitute legal advice and should not be relied upon as such. All cases, statutes, and sources cited are publicly available through official publications and reputable outlets. Readers should consult professional counsel for specific legal or compliance questions related to AI use.

See also: Data Provenance Emerges as Legal AI’s New Standard of Care