artificial intelligence/AI Archives

artificial intelligence/AI

Headnote to self: Third Circuit rejects fair use defense for AI training

by Hannah Hurley | Oct 15, 2026 | Copyrights

The US Court of Appeals for the Third Circuit affirmed a grant of partial summary judgment in favor of Thomson Reuters, finding that Westlaw’s editorial headnotes were copyrightable and that use of those headnotes to train an artificial intelligence (AI) legal-research platform did not constitute fair use. Characterizing the dispute as “no more than an ordinary copyright case,” the Court emphasized that its decision did not resolve broader questions concerning generative AI training. Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc., Case No. 25-2153 (3d Cir. Sept. 29, 2026) (Restrepo, Montgomery-Reeves, Bove, JJ.)

Thomson Reuters operates Westlaw, a legal-research platform that provides judicial opinions accompanied by editor-written headnotes summarizing key points of law. Although judicial opinions are not copyrightable, Thomson Reuters claimed copyright protection in the headnotes, which its editors prepare under detailed guidelines requiring them to identify and summarize significant legal principles in concise standalone statements.

ROSS developed a non-generative-AI legal-research platform designed to respond to users’ legal questions by identifying relevant passages from about 10 million judicial opinions. To train its platform, ROSS retained LegalEase Solutions to prepare about 25,000 memoranda, each containing a legal question and four to six responsive judicial opinions. LegalEase used thousands of Westlaw headnotes to formulate the questions. ROSS intended to compete directly with Westlaw and marketed its platform at comparable prices.

Thomson Reuters sued ROSS for copyright infringement. The district court granted partial summary judgment, finding that 2,243 headnotes identified by Thomson Reuters’ expert were sufficiently original to warrant copyright protection and that ROSS’s use of those headnotes was not fair use. The district court certified the originality and fair use issues for interlocutory appeal.

The Third Circuit first addressed copyrightability. Applying the “extremely low” originality threshold established in the Supreme Court’s 1991 Feist Publications, Inc. v. Rural Telephone Service Co. decision, the Third Circuit concluded that the headnotes contained sufficient creativity because editors exercised judgment in selecting which legal principles to summarize and how to express them. The Third Circuit rejected ROSS’s argument that copyright protection would effectively grant Thomson Reuters a monopoly over the law, explaining that judicial opinions remained freely available. The Court also rejected ROSS’s merger doctrine argument, finding that the underlying legal principles could be expressed in many ways.

Turning to fair use, the Third Circuit considered the four statutory factors under 17 U.S.C. § 107:

As to the purpose and character of the use, the Court found ROSS’s use commercial and, at most, minimally transformative because both platforms served the same purpose, which was helping users locate relevant legal materials. Although ROSS used the headnotes to train its AI platform rather than displaying them, it could have relied on freely available judicial opinions instead. This factor weighed against fair use.
The nature of the copyrighted work weighed slightly in ROSS’s favor because the headnotes primarily conveyed factual information and legal principles. The Court noted, however, that this factor generally carries limited weight.
The amount and substantiality of the portion used weighed against fair use [...]

Continue Reading

DMCA requirements for CMI removal not always met in AI code writing

by Jodi Benassi | Oct 1, 2026 | Copyrights

The US Court of Appeals for the Ninth Circuit affirmed the dismissal of Digital Millennium Copyright Act (DMCA) claims that alleged artificial intelligence (AI) tools removed or altered copyright management information (CMI) from the programmers’ protected works. Doe et al. v. GitHub, Inc., et al., Case No. 24-7700 (9th Cir. Sept. 16, 2026) (Thomas, Miller, Blumenfeld, JJ.)

GitHub operates a platform where developers can store, manage, and share software code. GitHub Copilot and OpenAI Codex are generative AI tools trained on millions of software projects available on GitHub. Copilot uses statistical patterns learned from its training data to generate code in response to user prompts. The plaintiffs are programmers who published copyrighted code in public GitHub repositories under open-source licenses that generally required attribution, including the author’s name and copyright notice.

The plaintiffs sued GitHub, Microsoft, and OpenAI, alleging that Copilot sometimes reproduced their code without the attribution, copyright notices, or license terms accompanying the code in the GitHub repositories. The plaintiffs asserted that these omissions violated § 1202(b) of the DMCA.

The district court dismissed the DMCA claim, reasoning that § 1202(b) required the allegedly infringing work to be identical to the original work from which the CMI had been removed. Because the plaintiffs alleged that Copilot generated near-identical, modified, or functionally equivalent versions of their code, the district court concluded that they had failed to state a claim. It certified for interlocutory appeal the question of whether § 1202(b) imposes an identicality requirement.

Removal or alteration of CMI

Section 1202(b) prohibits intentionally removing or altering CMI, and distributing works or copies of works knowing that CMI has been removed or altered without authority.

The plaintiffs argued that § 1202(b) does not require the defendant’s output to be literally identical to the copyrighted work. They contended that a literal-identicality rule would allow a defendant to evade the DMCA simply by making a trivial change to a copied work after removing its CMI – for example, changing one word on a page. The defendants, while conceding that literal identicality was not required, argued that § 1202(b) requires CMI to have been removed or altered from a copy of the plaintiff’s existing work. If Copilot instead generates a new or derivative work that never contained the plaintiff’s CMI, there is nothing from which CMI was “removed” or “altered.”

The Ninth Circuit rejected a literal identicality requirement and characterized “identicality” as a “misnomer.” The relevant inquiry is not whether the works are identical, but whether CMI was actually removed or altered from a copy of an existing protected work. Identicality may be evidence of removal where two works are otherwise identical, but the allegedly infringing version omits CMI contained in the original. Under those circumstances, a factfinder may reasonably infer that the CMI was removed. But literal identity is not required. Minor cosmetic changes will not necessarily defeat a claim where a defendant substantially or entirely reproduces an existing work and removes its CMI.

Applying that standard, the Ninth Circuit concluded that [...]

Continue Reading

Hallucinated help: USPTO disciplines patent attorney for failing to verify AI-generated citations

by Taylor MacDonald | Sep 24, 2026 | Patents

The United States Patent and Trademark Office (USPTO) issued a final order approving a settlement between its Office of Enrollment and Discipline (OED) and a registered patent attorney accused of professional misconduct arising from the use of artificial intelligence (AI) in patent litigation. In re Brian E. Mitchell, Proceeding No. D2026-16 (Director July 27, 2026).

Brian Mitchell, a registered patent attorney who had never appeared before the USPTO, used AI to assist in preparing a proposed claim construction chart in a patent case pending in the US District Court for the Eastern District of Pennsylvania. The AI-generated material included several incorrect or fabricated citations to the patent’s intrinsic record.

After the errors were identified, Mitchell corrected them within a day and cooperated fully with OED’s investigation. The final order noted that Mitchell “acknowledged his errors, demonstrated genuine contrition, and accepted responsibility,” and that he had no prior disciplinary history before the USPTO or other tribunals. The district court declined to impose sanctions.

Mitchell nevertheless stipulated that his conduct violated USPTO Rules of Professional Conduct governing competent representation, reasonable diligence, misrepresentation, and conduct prejudicial to the administration of justice. According to the order, the violations arose from Mitchell’s failure to appreciate AI’s limitations (including its propensity to hallucinate) and his failure to verify the AI-generated content before submitting it to the court.

As part of the settlement, Mitchell agreed to a public reprimand, publication of the final order in OED’s electronic Freedom of Information Act Reading Room, and publication of a Notice of Public Reprimand in the Official Gazette. The notice specifically warns practitioners that AI-generated errors “are not limited to extrinsic sources such as statutes, regulations, and case law,” but may also extend to intrinsic evidence.

The order further cautioned that the disciplinary proceeding may be considered as an aggravating factor, or as rebuttal evidence, in any future USPTO disciplinary matter involving Mitchell.

Picture this: AI enhancements to trademark search and examination

by Amit Kumar Saha, PhD | Apr 16, 2026 | Trademarks

The United States Patent and Trademark Office (USPTO) has announced several new artificial intelligence (AI) enhancements to its trademark search system and to the Trademark Center, continuing its broader effort to modernize trademark examination and improve both searchability and application quality.

One of the most notable updates is a new beta image-search capability within the USPTO’s trademark search system. This feature allows users to upload an image to identify similar marks with related design elements, functioning in a manner comparable to reverse image search tools available on commercial platforms. By enabling users to search for visually similar marks more intuitively, the tool may enhance clearance efforts, particularly for design marks that can be difficult to capture through traditional keyword-based searching.

The USPTO also announced that, beginning April 23, the Trademark Center will include a mark description and color claim generator. This feature is intended to assist applicants in drafting accurate and complete mark descriptions and color claims, which are often a source of inconsistency and procedural deficiency in trademark applications. By reducing guesswork and standardizing how such information is presented, the generator may help minimize office actions and improve overall application quality.

In addition to front-end user tools, the USPTO highlighted its Trademark Classification Agentic Codification Tool (Class ACT), an AI-driven system designed to automate certain back-end classification and coding functions. Class ACT assigns international classes to applications and generates design search codes and pseudo marks that make trademark records more easily searchable. Historically, these classification and coding steps could take months to complete, delaying examination and limiting searchability in the interim. The USPTO reports that Class ACT can perform these functions almost immediately, while still subject to human review to ensure accuracy and consistency.

Taken together, these developments reflect a continued shift toward integrating AI into both applicant facing and internal aspects of trademark prosecution. By accelerating classification, expanding search functionality, and assisting with application drafting, the USPTO seeks to streamline workflows while maintaining examiner oversight.

Practice note: Trademark applicants and practitioners may wish to incorporate the tools discussed above into clearance and filing strategies, particularly for design marks and applications involving complex descriptions or color claims. At the same time, AI generated results may surface a broader range of potentially relevant marks and classifications, underscoring the need for careful professional review. These tools are best viewed as supplements to – not substitutes for – informed legal analysis and strategic judgment.

TRAIN Act targets transparency in generative AI training practices

by Jolie B. Schenerman | Feb 12, 2026 | Copyrights

Representatives Madeleine Dean (D-PA) and Nathaniel Moran (R-TX) introduced a bipartisan bill entitled the Transparency and Responsibility for Artificial Intelligence Networks (TRAIN) Act in the US House of Representatives. The legislative objective is to provide a mechanism to help musicians, artists, writers, and other creators determine if their copyrighted work was used to train generative artificial intelligence (AI) models without their permission, and if so, to pursue compensation for that use. A bipartisan group of senators also reintroduced the TRAIN Act in the US Senate.

The proposed legislation would create a new legal mechanism allowing copyright owners to use federal court subpoena power to obtain information about the materials used to train generative AI models. The bill would establish the first federal statutory definition of “generative AI models.” A generative AI model would be defined as an AI model that “emulates the structure and characteristics of input data in order to generate derived synthetic content,” including images, video, audio, text, and other digital content, as well as any subsequent variation of such a model.

Under the proposed framework, an owner of one or more copyrighted works would be entitled to seek a subpoena compelling a generative AI developer to produce copies of, or records sufficient to identify, the copyrighted works (or portions thereof) that were used to train the developer’s model. To obtain a subpoena, the copyright owner would submit a request to a US district court clerk that includes a proposed subpoena and a sworn affidavit attesting to the owner’s subjective good faith belief that the developer used some or all of the owner’s copyrighted works to train its generative AI model. The affidavit would also attest that the subpoena’s purpose is to obtain the owner’s copyrighted materials or records to sufficiently identify the owner’s copyrighted materials used to train the model, and that the copies or records requested would be used solely to protect the copyright owner’s rights.

Once a subpoena was issued and served in accordance with the Federal Rules of Civil Procedure, the developer would be required to “expeditiously disclose” the requested copies or records to the copyright owner or an authorized representative.

The TRAIN Act also includes two enforcement provisions. First, if a developer fails to comply with a subpoena, the court may apply a rebuttable presumption that the developer did in fact use the copyrighted works to train its model. Second, if a copyright owner seeks a subpoena in bad faith, the recipient may pursue sanctions under Federal Rule of Civil Procedure 11.

If enacted, the TRAIN Act would significantly expand transparency obligations for developers of generative AI models and provide copyright owners with a new tool to investigate potential unauthorized uses of their works.

Precedential shift: USPTO clarifies patentability of AI training methods

by Hannah Hurley | Nov 20, 2025 | Patents

On November 4, 2025, the Director of the United States Patent and Trademark Office (USPTO) designated as precedential an appeals review panel (ARP) decision vacating the Patent Trial & Appeal Board’s § 101 rejection of claims directed to training machine learning models. Ex parte Desjardins, Appeal No. 24-000567 (ARP Sept. 26, 2025) (precedential).

The Board had previously concluded that claims covering continual learning techniques (such as adjusting model parameters to maintain performance across sequential tasks) were directed to an unpatentable abstract idea. The ARP, which included the USPTO Director, reversed that determination, holding that the claims integrated the abstract concept into a practical application by improving the functioning of machine learning models themselves. However, the ARP still rejected the claims under § 103 for obviousness.

Key takeaways

Technical improvements matter. Artificial intelligence (AI)-related inventions can satisfy Alice Step 2A when they demonstrate technical improvements, such as mitigating catastrophic forgetting and reducing storage complexity.
No blanket exclusion. The opinion cautions against categorically excluding AI innovations under § 101 and emphasizes that §§ 102, 103, and 112 remain the proper tools for assessing patent scope.
Precedential impact. The decision signals the USPTO’s commitment to aligning examination practices with US Court of Appeals for the Federal Circuit precedent while fostering innovation in AI and machine learning.

Practice note: For applicants, this precedential designation underscores the importance of framing AI-related claims around specific technical improvements rather than abstract concepts, which can be pivotal in overcoming § 101 challenges.

Fair use or foul play? The AI fair use copyright line

by Jodi Benassi | Jul 2, 2025 | Copyrights

The US District Court for the Northern District of California granted summary judgment in favor of an artificial intelligence (AI) company, finding that its use of lawfully acquired copyrighted materials for training and its digitization of acquired print works fell within the bounds of fair use. However, the district court explicitly rejected the AI company’s attempt to invoke fair use as a defense to rely on pirated copies of copyrighted works as lawful training data. Andrea Bartz, et al. v. Anthropic PBC, Case No. 24-CV-05417-WHA (N.D. Cal. June 23, 2025) (Alsup, J.)

Anthropic, an AI company, acquired more than seven million copyrighted books without authorization by downloading them from pirate websites. It also lawfully purchased print books, removed their bindings, scanned each page, and stored them in digitized, searchable files. The goal was twofold:

To create a central digital library intended, in Anthropic’s words, to contain “all the books in the world” and to be preserved indefinitely.
To use this library to train the large language models (LLMs) that power Anthropic’s AI assistant, Claude.

Each work selected for training the LLM was copied through four main stages:

Each selected book was copied from the library to create a working copy for training.
Each book was “cleaned” by removing low-value or repetitive content (e.g., footers).
Cleaned books were converted into “tokenized” versions by being simplified and split into short character sequences, then translated into numerical tokens using Anthropic’s custom dictionary. These tokens were repeatedly used in training, allowing the model to discover statistical relationships across massive text data.
Each fully trained LLM itself retained “compressed” copies of the books.

Once the LLM was trained, it did not output any of the books through Claude to the public. The company placed particular value on books with well-curated facts, structured analyses, and compelling narratives (i.e., works that reflected well-written creative expressions) because Claude’s users expected clear, accurate, and well-written responses to their questions.

Andrea Bartz, along with two other authors whose books were copied from pirated and purchased sources and used to train Claude, sued Anthropic for copyright infringement. In response, Anthropic filed an early motion for summary judgment on fair use only under Section 107 of the Copyright Act.

To assess the applicability of the fair use defense, the court separated and analyzed Anthropic’s actions across three distinct categories of use.

Transformative training (fair use)

The authors challenged only the inputs used to train the LLMs, not their outputs. The district court found that Anthropic’s use of copyrighted books to train its LLMs was a transformative use, comparable to how humans read and learn from texts and produce new, original writing. While the authors claimed that the LLMs memorized their creative expression, there was no evidence that Claude released infringing material to the public. The court concluded that using the works as training inputs – not for direct replication, but to enable the generation of new content – favored a finding of fair use.

Format-shifting copies (fair use)

[...]

Continue Reading

Copyright, AI, and Politics

by Paul Devinsky | May 22, 2025 | Copyrights

In early 2023, the US Copyright Office (CO) initiated an examination of copyright law and policy issues raised by artificial intelligence (AI), including the scope of copyright in AI-generated works and the use of copyrighted materials in AI training. Since then, the CO has issued the first two installments of a three-part report: part one on digital replicas, and part two on copyrightability.

On May 9, 2025, the CO released a pre-publication version of the third and final part of its report on Generative AI (GenAI) training. The report addresses stakeholder concerns and offers the CO’s interpretation of copyright’s fair use doctrine in the context of GenAI.

GenAI training involves using algorithms to train models on large datasets to generate new content. This process allows models to learn patterns and structures from existing data and then create new text, images, audio, or other forms of content. The use of copyrighted materials to train GenAI models raises complex copyright issues, particularly issues arising under the “fair use” doctrine. The key question is whether using copyrighted works to train AI without explicit permission from the rights holders is fair use and therefore not an infringement or whether such use violates copyright.

The 107-page report provides a thorough technical and legal overview and takes a carefully calculated approach responding to the legal issues underlying fair use in GenAI. The report suggests that each case is context specific and requires a thorough evaluation of the four factors outlined in Section 107 of the Copyright Act:

The purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes
The nature of the copyrighted work
The amount and substantiality of the portion used in relation to the copyrighted work as a whole
The effect of the use upon the potential market for or value of the copyrighted work.

With regard to the first factor, the report concludes that GenAI training run on large diverse datasets “will often be transformative.” However, the use of copyright-protected materials for AI model training alone is insufficient to justify fair use. The report states that “transformativeness is a matter of degree of the model and how it is deployed.”

The report notes that training a model is most transformative where “the purpose is to deploy it for research, or in a closed system that constrains it to a non-substitutive task,” as opposed to instances where the AI output closely tracks the creative intent of the input (e.g., generating art, music, or writing in a similar style or substance to the original source materials).

As to the second factor (commercial nature of the use), the report notes that a GenAI model is often the product of efforts undertaken by distinct and multiple actors, some of which are commercial entities and some of which are not, and that it is typically difficult to discern attribution and definitively determine that a model is the product of a commercial or a noncommercial [...]

Continue Reading

Human Authorship Required: AI Isn’t an Author Under Copyright Act

by Jodi Benassi | Mar 27, 2025 | Copyrights

The US Court of Appeals for the District of Columbia upheld a district court ruling that affirmed the US Copyright Office’s (CO) denial of a copyright application for artwork created by artificial intelligence (AI), reaffirming that human authorship is necessary for copyright registration. Thaler v. Perlmutter, Case No. 23-5233 (D.C. Cir. Mar. 18, 2025) (Millett, Wilkins, Rogers, JJ.)

Stephen Thaler, PhD, created a generative AI system that he named the Creativity Machine. The machine created a picture that Thaler titled, “A Recent Entrance to Paradise.” Thaler applied to the CO for copyright registration for the artwork, listing the Creativity Machine as the author and Thaler as the copyright owner.

The CO denied Thaler’s application because “a human being did not create the work.” Thaler twice sought reconsideration of the application, which the CO denied because the work lacked human authorship. Thaler subsequently sought review in the US District Court for the District of Columbia, which affirmed the CO’s denial of registration. The district court concluded that “[h]uman authorship is a bedrock requirement of copyright.” Thaler appealed.

The DC Circuit reaffirmed that the Creativity Machine could not be considered the author of a copyrighted work. The Copyright Act of 1976 mandates that to be eligible for copyright, a work must be initially authored by a human being. The Court highlighted key provisions of the Copyright Act that only make sense if “author” is interpreted as referring to a human being. For instance:

A copyright is a property right that immediately vests in the author. Since AI cannot own property, it cannot hold copyright.
Copyright protection lasts for the author’s lifetime, but machines do not have lifespans.
Copyright is inheritable, but machines have no surviving spouses or heirs.
Transferring a copyright requires a signature, and machines cannot provide signatures.
Authors of unpublished works are protected regardless of their nationality or domicile, yet machines do not have a domicile or national identity.
Authors have intentions, but machines lack consciousness and cannot form intentions.

The DC Circuit concluded that the statutory provisions, as a whole, make human activity a necessary condition for authorship under the Copyright Act.

The DC Circuit noted that the human authorship requirement is not new, referencing multiple judicial decisions, including those from the Seventh and Ninth Circuits, where appellate courts have consistently ruled that authors must be human.

Practice Note: Only humans, not their tools, can author copyrightable works of art. Images autonomously generated are not eligible for copyright. However, works created by humans who used AI are eligible for copyright depending on the circumstances, how the AI tool operates, and to what degree the AI tool was used to create the final work. Authors whose works are assisted by AI should seek advice of counsel to determine whether their works are copyrightable.

Assessing Inputs: Determining AI’s Role in US Intellectual Property Protections

by Anisa Noorassa | Feb 6, 2025 | Copyrights

The US Patent & Trademark Office (PTO) issued additional guidance on the contribution of artificial intelligence (AI) in its January 2025 AI Strategy. Similarly, the US Copyright Office issued part two of its “Copyright and Artificial Intelligence” report, addressing the copyrightability of AI- or partially AI-made works. Both agencies appear to be walking a fine line by accepting that AI has become increasingly pervasive while maintaining human contribution requirements for protected works and inventions.

In its published strategy, the PTO states that its vision is to unleash “America’s potential through the adoption of AI.” The strategy describes five focus areas:

Advancing the development of intellectual property policies that promote inclusive AI innovation and creativity.
Building best-in-class AI capabilities by investing in computational infrastructure, data resources, and business-driven product development.
Promoting the responsible use of AI within the PTO and across the broader innovation ecosystem.
Developing AI expertise within the PTO’s workforce.
Collaborating with other US government agencies, international partners, and the public on shared AI priorities.

The PTO stated that it is still evaluating the issue of AI-assisted inventions but reaffirmed its February 2024 guidance on inventorship for AI-assisted inventions. That guidance indicates that while AI-assisted inventions are not categorically unpatentable, the inventorship analysis should focus on human contributions.

Likewise, the Copyright Office discussed public comments regarding AI contributions to copyright, weighing the benefits of AI in assisting and empowering creators with disabilities against the harm to artists working to make a living. Ultimately the Copyright Office affirmed that AI, when used as a tool, can generate copyrightable works only where a human is able to determine the expressive elements contained in the work. The Copyright Office stated that creativity in the AI prompt alone is, at this state, insufficient to satisfy the human expressive input required to produce a copyrightable work.

Older Posts »

BLOG EDITORS

Paul Devinsky

Amol Parikh

Jodi Benassi

BLOG EDITORS

STAY CONNECTED

TOPICS

ARCHIVES

RECENT POSTS