As artificial intelligence (AI) continues to reshape innovation, creativity, and business operations, United States courts issued several landmark decisions in 2025 and early 2026 that clarify the boundaries of intellectual property protection in the AI era. These rulings—updated with key procedural developments through late May 2026—provide much-needed clarity for companies using generative AI and for creators whose works may be feeding those systems, helping you protect your competitive edge in 2026 and beyond.
At Sherman IP, we closely monitor these developments to help our clients proactively safeguard their portfolios and mitigate risks. Below is a summary of the most significant recent case law, along with practical implications for your business.
Supreme Court Denies Certiorari in Thaler v. Perlmutter – AI Cannot Be an Author (or Inventor)
In a closely watched appeal, the U.S. Supreme Court on March 2, 2026, declined to review Thaler v. Perlmutter (Case No. 25-449), letting stand the D.C. Circuit’s 2025 affirmance of the U.S. Copyright Office’s refusal to register a purely AI-generated artwork as copyrightable.
Dr. Stephen Thaler sought copyright protection for the image “A Recent Entrance to Paradise,” created autonomously by his AI system DABUS, listing the AI (not himself) as author. The Copyright Office, district court, and D.C. Circuit all held that the U.S. Copyright Act requires human authorship—a “bedrock requirement.” The Supreme Court’s denial of certiorari effectively ends this chapter: under current law, AI-generated works lacking sufficient human creative input are not eligible for copyright protection.
In practice, this means a one-line prompt like “create a landscape” will not qualify, but a detailed prompt combined with substantial human selection, editing, and arrangement of multiple AI outputs likely will qualify for copyright protection.
Parallel Patent Implications: Thaler’s related patent applications listing DABUS as inventor were similarly rejected. The United States Patent and Trademark Office’s (USPTO) November 2025 revised inventorship guidance reinforces that only natural persons can be inventors; AI is treated as a sophisticated tool, and “conception” remains an inherently human act. Applicants must document specific human contributions to AI-assisted inventions to secure patent protection. For a deeper discussion of how to build and preserve that record, see our companion article, AI in Innovation: Documenting Human Contributions for Patent Success.
Takeaway for Clients: Purely AI-generated outputs (e.g., images, text or other content created with minimal human prompting or editing) generally are not eligible for copyright protection. Likewise, AI cannot be named as an inventor on a patent application, and patent rights require at least one human inventor who made a significant contribution to the claimed invention. However, works and inventions involving meaningful human creativity or inventive contribution may qualify for protection. We recommend maintaining detailed records of human involvement to support future registration, patent prosecution, or enforcement efforts.
First Substantive Fair Use Rulings on AI Training Data
2025 saw the first district court merits decisions addressing whether using copyrighted works to train generative AI models constitutes fair use under 17 U.S.C. § 107. These rulings draw important distinctions based on the source of training data and the nature of the AI application. May 2026 also marked an important procedural milestone, as one of the earliest AI copyright class-action settlements received final approval.
Bartz v. Anthropic (N.D. Cal., June 23, 2025). Judge William Alsup granted partial summary judgment to Anthropic, holding that training its Claude LLMs on lawfully acquired books was “exceedingly” and “spectacularly” transformative fair use. The court analogized the process to a human reading books to learn patterns rather than reproducing expressive content. However, Anthropic’s separate act of downloading millions of pirated books from shadow libraries was not fair use.
- May 2026 Update: The parties reached a historic approximately $1.5 billion class settlement (roughly $3,000 per infringed work). The final fairness hearing was held on May 14, 2026. The proceeding was smooth with “little drama”: the claims rate was exceptionally high (~91.3%), opt-outs and objections were minimal, and the court took the matter under submission after requesting a short supplemental brief from Anthropic (due May 21) addressing a handful of late opt-outs. Final approval is expected imminently—potentially within days or weeks. Payments to class members will follow after any appeals period. The core fair-use holding remains undisturbed.
Kadrey v. Meta (N.D. Cal., June 25, 2025). Judge Vince Chhabria granted Meta summary judgment, finding training its Llama models on copyrighted books (including those from shadow libraries) fair use on the specific record presented. The decision emphasized the transformative nature of the use but cautioned that market-harm evidence could tip the fourth fair use factor (the effect of the use on the potential market for or value of the copyrighted work) against defendants in future cases. The case remains ongoing following the March 2026 amendment to the complaint.
Thomson Reuters v. Ross Intelligence (D. Del., Feb. 11, 2025). In contrast, Judge Stephanos Bibas (sitting by designation) held that Ross’s use of Thomson Reuters’ copyrighted Westlaw headnotes to train a competing AI legal research tool was not fair use. The court found the use non-transformative and directly substitutive of the plaintiff’s market. The decision is on interlocutory appeal to the Third Circuit, with oral argument scheduled for June 11, 2026.
While these early wins favor AI developers when data is lawfully acquired, several high-profile cases (including NYT v. OpenAI, Getty Images v. Stability AI, and music-industry lawsuits) remain pending or have resolved through licensing deals—signaling that the final chapter on market-harm analysis is still being written.
Practical Advice for Clients
These developments signal a maturing—but still evolving—legal framework. Here are concrete steps you can take right now:
- AI Developers & Users
- Prioritize licensed or publicly available datasets; document every human contribution (prompt engineering, iterative editing, output curation).
- Implement output filters and “human-in-the-loop” review protocols to minimize infringement risk.
- Content Owners & Creators
- Review licensing opportunities with major AI platforms—many are now proactively negotiating bulk deals post-Bartz.
- Consider technical protection (watermarking, metadata tagging) and monitor for unauthorized scraping.
- All Clients
- Schedule a portfolio audit with your Sherman IP attorney to stress-test AI exposure in your current workflows.
KEY TAKEAWAYS
- Pure AI outputs = no copyright.
- Lawful-data training = strong fair-use defense (with important caveats regarding source of data).
- Document human creativity.
- License where possible.
- Bartz class action settlement is now moving rapidly toward final approval—further accelerating licensing momentum across the industry.
Sherman IP remains at the forefront of AI-IP strategy. Our team is available to review your specific use cases, assist with registrations, negotiate licenses, or defend/enforce rights in this rapidly changing landscape.
Please contact your Sherman IP attorney or reach out to us at info@shermanip.com to schedule a consultation. We look forward to partnering with you to turn these legal developments into competitive advantages.
Best Regards,
The Sherman IP LLP Team
Protecting Your Vision, Securing Your Future
Take advantage of a FREE consultation with one of our expert IP attorneys today, who will evaluate your matterand provide a proposed strategy to proceed.