{"id":826,"date":"2026-05-28T04:44:09","date_gmt":"2026-05-28T04:44:09","guid":{"rendered":"https:\/\/www.perspireip.com\/blog\/ai-and-copyright-2026-legal-shifts\/"},"modified":"2026-05-28T04:52:04","modified_gmt":"2026-05-28T04:52:04","slug":"ai-and-copyright-2026-legal-shifts","status":"publish","type":"post","link":"https:\/\/www.perspireip.com\/blog\/ai-and-copyright-2026-legal-shifts\/","title":{"rendered":"AI and Copyright in 2026: 5 Critical Legal Shifts Every IP Counsel Must Track"},"content":{"rendered":"\n<p>The legal ground under <strong>AI and copyright<\/strong> is shifting faster than most IP teams can keep up with. If you advise creators, manage a brand portfolio, or build AI-assisted tools, the rules you operated under in 2023 are not the rules that apply in 2026 \u2014 and the gap is already costing money for rights holders, model developers, and the lawyers caught in the middle.<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/www.perspireip.com\/blog\/wp-content\/plugins\/perspire-images\/featured-28.jpg\" alt=\"AI and copyright 2026 legal landscape\" \/><\/figure>\n\n\n\n\n\n<p>Some questions are now settled. AI systems cannot, on their own, be authors of copyrighted works in the United States. The D.C. Circuit confirmed that in <em>Thaler v. Perlmutter<\/em> in 2025, and the Supreme Court declined to review the decision in March 2026. That door is closed. But almost everything else \u2014 fair use in model training, the threshold for human authorship in AI-assisted works, and the scope of statutory damages in infringement actions against AI companies \u2014 remains contested, and dozens of cases are heading toward dispositive rulings this year.<\/p>\n\n\n\n<p>This guide unpacks where <strong>AI and copyright<\/strong> law actually stands in 2026, what the U.S. Copyright Office&#8217;s recent guidance means in practice, where the litigation surge is heading, and what rights holders, inventors, and counsel should be doing right now to stay ahead of a regime that is being written in real time.<\/p>\n\n\n\n<div class=\"wp-block-rank-math-toc-block\" id=\"rank-math-toc\"><h2>Table of Contents<\/h2><nav><ul><li class=\"\"><a href=\"#ai-and-copyright-what-is-settled\">AI and Copyright: What Is Actually Settled in 2026<\/a><\/li><li class=\"\"><a href=\"#ai-and-copyright-training-fair-use\">AI Training, Fair Use, and the Copyright Office&#8217;s 2025 Guidance<\/a><\/li><li class=\"\"><a href=\"#ai-and-copyright-human-authorship-threshold\">The Human Authorship Threshold: Where Counsel Earns Its Value<\/a><\/li><li class=\"\"><a href=\"#ai-and-copyright-litigation-surge\">Litigation Is Accelerating: What&#8217;s on the Docket in 2026<\/a><\/li><li class=\"\"><a href=\"#ai-and-copyright-licensing-standard\">Licensing Is Becoming the Default, Not the Exception<\/a><\/li><li class=\"\"><a href=\"#ai-and-copyright-what-to-do-now\">What This Means for Your Clients: Practical Moves in 2026<\/a><\/li><li class=\"\"><a href=\"#perspireip-ai-copyright\">How PerspireIP Supports AI and Copyright Strategy<\/a><\/li><li class=\"\"><a href=\"#conclusion\">Final Thoughts on AI and Copyright in 2026<\/a><\/li><li class=\"\"><a href=\"#faq-ai-and-copyright\">Frequently Asked Questions About AI and Copyright<\/a><ul><li class=\"\"><a href=\"#can-an-ai-system-be-listed-as-the-author-of-a-copyrighted-work-in-the-u-s\">Can an AI system be listed as the author of a copyrighted work in the U.S.?<\/a><\/li><li class=\"\"><a href=\"#is-training-a-generative-ai-model-on-copyrighted-works-fair-use\">Is training a generative AI model on copyrighted works fair use?<\/a><\/li><li class=\"\"><a href=\"#are-ai-assisted-works-copyrightable\">Are AI-assisted works copyrightable?<\/a><\/li><li class=\"\"><a href=\"#what-is-the-statutory-damages-threshold-for-an-ai-copyright-infringement-claim\">What is the statutory damages threshold for an AI copyright infringement claim?<\/a><\/li><li class=\"\"><a href=\"#should-ai-companies-license-training-data-even-if-fair-use-might-apply\">Should AI companies license training data even if fair use might apply?<\/a><\/li><\/ul><\/li><\/ul><\/nav><\/div>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ai-and-copyright-what-is-settled\">AI and Copyright: What Is Actually Settled in 2026<\/h2>\n\n\n\n<p>Start with the bedrock. The U.S. Copyright Office has been unambiguous, repeatedly, that human authorship is a statutory prerequisite for copyright protection. The Office&#8217;s guidance, reaffirmed across multiple registration policy statements and culminating in its 2025 Part 2 report on copyright and AI, treats outputs generated by an AI system without sufficient human creative control as ineligible for registration.<\/p>\n\n\n\n<p>The courts have backed that view. In <a href=\"https:\/\/www.cadc.uscourts.gov\/\" target=\"_blank\" rel=\"noopener\">Thaler v. Perlmutter<\/a>, computer scientist Stephen Thaler sought to register a work generated by an AI system he called the &#8220;Creativity Machine,&#8221; listing the machine as the author. The Copyright Office refused. The D.C. district court affirmed. The D.C. Circuit affirmed in 2025. And in March 2026 the Supreme Court declined to take up the case, leaving the lower courts&#8217; rulings as the controlling law: AI, standing alone, cannot be an author for copyright purposes.<\/p>\n\n\n\n<p>That is the floor. What sits on top of it \u2014 the question every working creator and counsel cares about \u2014 is where the human authorship line gets drawn when humans and AI tools collaborate. And on that question, the law is still being written.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ai-and-copyright-training-fair-use\">AI Training, Fair Use, and the Copyright Office&#8217;s 2025 Guidance<\/h2>\n\n\n\n<p>The most economically significant <strong>AI and copyright<\/strong> question of 2026 is whether training a generative model on copyrighted works without a license constitutes fair use. The Copyright Office weighed in with notable force in May 2025, when it published Part 3 of its multi-part report on copyright and artificial intelligence focused on generative AI training.<\/p>\n\n\n\n<p>The Office&#8217;s bottom line was carefully phrased but pointed: where AI developers ingest copyrighted works to train models that then produce content competing with the original works in their existing markets, that use is unlikely to qualify as fair use under 17 U.S.C. \u00a7 107. The four-factor analysis cuts against the developer when the output substitutes for the input \u2014 and even further when access to the training corpus was acquired through pirated sources rather than licensed datasets.<\/p>\n\n\n\n<p>The Office&#8217;s report is not binding on courts, but it carries persuasive weight. Federal judges have been citing it in early rulings, and amicus briefs are quoting it heavily. As <a href=\"https:\/\/builtin.com\/artificial-intelligence\/ai-copyright\" target=\"_blank\" rel=\"noopener\">industry coverage of the AI copyright fight<\/a> has noted, the report effectively moved the goalposts on the fair use defense that AI developers had been relying on since the launch of large language models.<\/p>\n\n\n\n<p>What does that mean operationally? Two things. First, &#8220;we used the open web&#8221; is no longer a credible defense if the training corpus included works behind paywalls, pirated dataset dumps, or material scraped in violation of terms of service. Second, model developers who can document a licensed training data pipeline have a meaningfully better posture than those who cannot \u2014 and the gap is widening with every new lawsuit.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ai-and-copyright-human-authorship-threshold\">The Human Authorship Threshold: Where Counsel Earns Its Value<\/h2>\n\n\n\n<p>Here is the part where I see clients get into the most trouble. The extremes of the human authorship doctrine are clear. A single text prompt that produces a finished image \u2014 no human selection, no editorial revision, no creative arrangement \u2014 does not produce a copyrightable work. The Copyright Office has refused to register dozens of such works since 2023, and the courts have not disturbed those refusals. <a href=\"https:\/\/garrettham.com\/ai-generated-works-copyright\/\" target=\"_blank\" rel=\"noopener\">Practitioner commentary<\/a> on the doctrine consistently reaches the same conclusion at the prompt-only end of the spectrum.<\/p>\n\n\n\n<p>On the other end, work that is substantially human-directed remains fully protectable. A novelist who drafts a manuscript by hand, uses AI as a spell-checker, and edits the result line by line still owns a copyright in the finished book. A studio that uses an AI tool to generate concept-art rough drafts, then layers human compositional, color, and lighting decisions on top, owns the final composite work.<\/p>\n\n\n\n<p>The murky territory \u2014 and it is enormous \u2014 is the middle. A photographer uses an AI tool to extend a background. A songwriter uses a generative system to suggest chord progressions and then re-records the result with live instrumentation. A software engineer accepts AI code completions in some functions and writes others by hand. In each case, the registrability of the final work depends on a careful, work-by-work analysis of what the human contributed and how much creative judgment was applied to the AI&#8217;s output.<\/p>\n\n\n\n<p>The Copyright Office&#8217;s current registration guidance requires applicants to disclose AI-generated material in their applications and to limit their registration claim to the human-authored portions. Counsel earns its value in helping clients structure their workflows so that the human creative contribution is documented contemporaneously and identifiable in the final work. If you&#8217;re advising AI-assisted creators, our <a href=\"https:\/\/www.perspireip.com\/blog\/copyright-vs-trademark-vs-patent-guide\/\">comparison of copyright, trademark, and patent protection<\/a> is a good companion read on which IP framework applies when.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ai-and-copyright-litigation-surge\">Litigation Is Accelerating: What&#8217;s on the Docket in 2026<\/h2>\n\n\n\n<p>The volume of <strong>AI and copyright<\/strong> litigation is unlike anything the IP bar has seen since the early file-sharing era. By the end of 2025, more than 70 copyright infringement lawsuits had been filed by rights holders against AI companies. That number has continued to climb through 2026. The plaintiffs include book publishers, newspapers, music labels, photographers, visual artists, and stock-image services. The defendants include essentially every major foundation-model developer.<\/p>\n\n\n\n<p>According to the <a href=\"https:\/\/copyrightalliance.org\/ai-copyright-lawsuit-developments-2025\/\" target=\"_blank\" rel=\"noopener\">Copyright Alliance&#8217;s tracking of 2025 developments<\/a>, the cases cluster around three issues: whether training constitutes infringement, whether outputs that reproduce identifiable portions of training inputs are themselves infringing, and whether secondary liability attaches to platforms that distribute infringing AI outputs to end users. Several cases survived motions to dismiss in 2025 and are now in discovery or summary judgment briefing.<\/p>\n\n\n\n<p>Practitioners watching the docket closely \u2014 including the <a href=\"https:\/\/www.clearygottlieb.com\/news-and-insights\/publication-listing\/the-open-questions-in-us-generative-ai-copyright-litigation\" target=\"_blank\" rel=\"noopener\">Cleary Gottlieb team tracking open questions in generative AI copyright litigation<\/a> \u2014 expect a series of dispositive rulings in 2026 that will start to answer three foundational questions. Does training on lawfully acquired copyrighted works qualify as fair use? Does the answer change when training corpus access was obtained through unauthorized scraping or pirated datasets? And what proof of substantial similarity is required for output-level infringement when the model &#8220;memorized&#8221; portions of the training data?<\/p>\n\n\n\n<p>Whatever the answers, the cumulative effect of so many parallel cases is going to reshape the commercial landscape. Settlements are already happening at significant numbers, and the disclosure regime that emerges from discovery in these cases will give the world its first detailed look at how foundation models were actually trained.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ai-and-copyright-licensing-standard\">Licensing Is Becoming the Default, Not the Exception<\/h2>\n\n\n\n<p>The most telling signal in the <strong>AI and copyright<\/strong> ecosystem is not what the courts are saying \u2014 it&#8217;s what the parties are doing. Over the last 18 months, formal licensing agreements between content owners and AI companies have moved from a curiosity to a standard practice. Major publishers have signed multi-year deals with the largest foundation-model developers. Stock-image houses have introduced AI training license tiers. News organizations have either signed up or sued; almost none are sitting still.<\/p>\n\n\n\n<p>This trend, tracked by firms like <a href=\"https:\/\/www.nortonrosefulbright.com\/en\/knowledge\/publications\/ce8eaa5f\/ai-in-litigation-series-an-update-on-ai-copyright-cases-in-2026\" target=\"_blank\" rel=\"noopener\">Norton Rose Fulbright<\/a>, matters for two reasons. First, it reduces the developer&#8217;s litigation exposure by replacing uncertain fair-use defenses with paid, contractual rights. Second, the very fact that license markets are clearing at scale undercuts the fair use argument \u2014 because one of the four statutory factors, the effect on the market for the original work, is harder to argue in your favor when an actual market for that use has emerged.<\/p>\n\n\n\n<p>For rights holders, the practical takeaway is that the door is open. Foundation-model developers who lawfully license training data are paying real money for it, and the rate cards keep firming up. For developers, the takeaway is that a licensed posture is increasingly the only defensible commercial posture \u2014 and the cost is now a known line item rather than an open-ended risk.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ai-and-copyright-what-to-do-now\">What This Means for Your Clients: Practical Moves in 2026<\/h2>\n\n\n\n<p>The 2026 <strong>AI and copyright<\/strong> environment rewards preparation and punishes drift. Three audiences should be acting now.<\/p>\n\n\n\n<p><strong>For rights holders<\/strong>, copyright registration is no longer optional housekeeping \u2014 it is the threshold for statutory damages and attorneys&#8217; fees in any enforcement action against an AI company. Without timely registration (before infringement or within three months of first publication), you are limited to actual damages and disgorged profits, which are notoriously hard to prove in AI training cases where the value flows through opaque model weights. Register early, register comprehensively, and keep registrations current as your catalog grows. Our <a href=\"https:\/\/www.perspireip.com\/blog\/startup-ip-portfolio-founders-build-plan\/\">guide to building a startup IP portfolio<\/a> walks through how to prioritize registration spend.<\/p>\n\n\n\n<p><strong>For inventors and technologists building AI-assisted tools<\/strong>, the protectability of your output depends on documented human creative control. Build the documentation into your workflow. Time-stamped drafts, decision logs, version histories, and contemporaneous notes on which AI suggestions were accepted, modified, or rejected are not just nice to have \u2014 they are the evidentiary record you will need if a registration is challenged or a downstream license is contested. Treat the AI as a tool whose use is logged, not a co-author whose contributions are invisible.<\/p>\n\n\n\n<p><strong>For counsel advising AI companies<\/strong>, the litigation exposure profile is now defined by three things: training data provenance, opt-out compliance, and licensing posture. If your client cannot document where its training corpus came from, whether it honored creator opt-outs, and how much of it is under license, you cannot accurately price its exposure. That has implications for fundraising disclosures, M&amp;A diligence, indemnification clauses in customer contracts, and insurance underwriting. The work to fix these gaps is unglamorous but urgent.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"perspireip-ai-copyright\">How PerspireIP Supports AI and Copyright Strategy<\/h2>\n\n\n\n<p>PerspireIP supports copyright registration strategy, portfolio protection, and IP counseling for creators, technologists, and the attorneys who advise them. Our team works on registration filings, AI-assisted work documentation frameworks, infringement detection across AI-generated outputs, and IP due diligence for AI companies and their investors. We collaborate with outside counsel on litigation support, prior art equivalents in the copyright context (chain of authorship, registration timing, derivative work analysis), and licensing diligence for both sides of the table.<\/p>\n\n\n\n<p>Where we earn our fees is in the operational layer \u2014 translating the rapidly evolving doctrine into workflows, registration plans, and documentation templates your team can actually use. The law is being written right now. Position your clients ahead of it. To see how we approach broader IP strategy, our <a href=\"https:\/\/www.perspireip.com\/blog\/ip-due-diligence-business-deal\/\">IP due diligence guide<\/a> covers what acquirers and investors are scrutinizing in 2026.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"conclusion\">Final Thoughts on AI and Copyright in 2026<\/h2>\n\n\n\n<p><strong>AI and copyright<\/strong> is the most active area of U.S. intellectual property law since the dawn of the internet. The settled questions \u2014 AI cannot be an author \u2014 are a small island in a much larger ocean of contested issues. Rights holders, AI developers, and the counsel who advise them are operating in a regime that is being defined case by case, ruling by ruling, license by license.<\/p>\n\n\n\n<p>The right move is not to wait for clarity. The right move is to act on the principles that are clear, document the activity that isn&#8217;t, and stay close to a counsel team that&#8217;s reading every ruling as it comes down. Reach out to the PerspireIP team and we&#8217;ll scope a copyright and AI strategy review tailored to your portfolio, your product, or your client base. The law is moving. So should you.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"faq-ai-and-copyright\">Frequently Asked Questions About AI and Copyright<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"can-an-ai-system-be-listed-as-the-author-of-a-copyrighted-work-in-the-u-s\">Can an AI system be listed as the author of a copyrighted work in the U.S.?<\/h3>\n\n\n\n<p>No. The U.S. Copyright Office requires human authorship, and the D.C. Circuit confirmed in <em>Thaler v. Perlmutter<\/em> (2025) that an AI cannot be listed as the author. The Supreme Court declined review in March 2026, leaving the rule in place.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"is-training-a-generative-ai-model-on-copyrighted-works-fair-use\">Is training a generative AI model on copyrighted works fair use?<\/h3>\n\n\n\n<p>Not categorically. The Copyright Office concluded in 2025 that training that produces outputs competing with the original works in their existing markets is unlikely to qualify as fair use. Several dispositive rulings in 2026 will test this position. The safer commercial posture is to license training data where feasible.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"are-ai-assisted-works-copyrightable\">Are AI-assisted works copyrightable?<\/h3>\n\n\n\n<p>Yes, when human creative control is substantial. The human-authored portions of an AI-assisted work are registrable; the AI-generated portions are not. Applicants must disclose AI-generated material in their registration applications. Documenting the human contribution contemporaneously is the practical best practice.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"what-is-the-statutory-damages-threshold-for-an-ai-copyright-infringement-claim\">What is the statutory damages threshold for an AI copyright infringement claim?<\/h3>\n\n\n\n<p>Statutory damages and attorneys&#8217; fees are available only if the work was registered before the infringement began, or within three months of first publication. Without timely registration, plaintiffs are limited to actual damages and disgorged profits \u2014 which are notoriously hard to prove when the alleged use is model training.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"should-ai-companies-license-training-data-even-if-fair-use-might-apply\">Should AI companies license training data even if fair use might apply?<\/h3>\n\n\n\n<p>Increasingly, yes. The growth of formal licensing markets makes the fair use defense harder to win on the market-effect factor, and licensed training data substantially reduces both litigation exposure and the diligence risk that surfaces in fundraising and M&amp;A. Most major foundation-model developers now have at least partial licensed pipelines.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>The legal ground under AI and copyright is shifting fast. Thaler v. Perlmutter, the May 2025 Copyright Office report, the litigation surge, and what counsel should do now.<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[58,3,55],"tags":[182,175,187,185,98,179,186,184,188,189,183],"class_list":["post-826","post","type-post","status-publish","format-standard","hentry","category-patent","category-ip-strategy","category-trademark","tag-ai-and-copyright","tag-ai-copyright-law","tag-ai-litigation","tag-ai-training","tag-copyright","tag-copyright-office","tag-copyright-registration","tag-fair-use","tag-generative-ai","tag-human-authorship","tag-thaler-v-perlmutter"],"_links":{"self":[{"href":"https:\/\/www.perspireip.com\/blog\/wp-json\/wp\/v2\/posts\/826","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.perspireip.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.perspireip.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.perspireip.com\/blog\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.perspireip.com\/blog\/wp-json\/wp\/v2\/comments?post=826"}],"version-history":[{"count":2,"href":"https:\/\/www.perspireip.com\/blog\/wp-json\/wp\/v2\/posts\/826\/revisions"}],"predecessor-version":[{"id":828,"href":"https:\/\/www.perspireip.com\/blog\/wp-json\/wp\/v2\/posts\/826\/revisions\/828"}],"wp:attachment":[{"href":"https:\/\/www.perspireip.com\/blog\/wp-json\/wp\/v2\/media?parent=826"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.perspireip.com\/blog\/wp-json\/wp\/v2\/categories?post=826"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.perspireip.com\/blog\/wp-json\/wp\/v2\/tags?post=826"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}