{"path":"/topics/ai-and-society/copyright-claims-are-cheap","content":{"title":"Copyright Claims Are Cheap. Challenging Them Isn't.","summary":"Why generative AI is unlikely to bring back copyright formalities in the way Cory Doctorow predicts","body":"In his October 31 essay, [*Formalities*](https://pluralistic.net/2026/10/31/burn-berne/), Cory Doctorow makes an intriguing prediction: generative AI is going to force corporations to take copyright administration seriously again.\n\nHis argument begins with a straightforward proposition about American copyright law. Copyright protects human creativity, not the outputs of autonomous machines. As businesses increasingly mix human and AI-generated work, they will therefore need to distinguish which parts of their products are actually protected.\n\nDoctorow concludes that businesses will be forced to maintain detailed records of how their creative products are made. He predicts a return to copyright formalities, this time imposed by the private sector rather than the government. Companies will have to document the provenance of every pixel, word and note if they want to defend the intellectual property they claim to own.\n\nThere is something appealing about this prediction. Copyright formalities fell out of favour partly because they imposed administrative burdens on rights holders, leaving the public with an enormous universe of works whose ownership and protection are difficult to establish. Perhaps AI will finally force corporations to confront those problems.\n\nBut I think Doctorow makes a significant leap from a correct observation about copyright doctrine to an unlikely prediction about corporate behaviour.\n\n**Copyright law does not operate in a vacuum. It operates within a market in which property rights are asserted, recognised, traded and valued, frequently without anyone testing their validity.**\n\nAnd that market has powerful reasons to preserve the existing arrangements.\n\n## The difference between having copyright and claiming copyright\n\nLet's begin with the legal point on which Doctorow is substantially correct.\n\nUnder current US law, purely AI-generated expression is not eligible for copyright protection. The Copyright Office's [January 2025 report](https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf) confirms that human authorship remains essential.\n\nThe report also makes an important qualification. Human creative contributions can remain protected even when a work includes AI-generated material. Editing, arranging or incorporating such material doesn't necessarily invalidate the copyright in the resulting work. Nor does using AI as an assistant automatically make a work uncopyrightable.\n\nThe DC Circuit's decision in [*Thaler v. Perlmutter*](https://law.justia.com/cases/federal/appellate-courts/cadc/23-5233/23-5233-2025-03-18.html) in March 2025 affirmed that a machine cannot be the sole author of a copyrighted work.\n\nNone of this is particularly helpful to a corporation hoping to claim exclusive rights over material generated entirely by AI.\n\nBut copyrightability is only one part of the question.\n\nSuppose a corporation publishes a successful illustrated book. It claims copyright in the ordinary way, identifying the corporation as the rights holder.\n\nPerhaps an employee drew the illustrations. Perhaps a contractor drew them and assigned the rights. Perhaps the employee generated them with an AI tool and made extensive creative modifications. Or perhaps the images came almost directly from an image generator.\n\nThese possibilities have different legal implications.\n\nBut from the outside, they might look exactly the same: a book bearing a corporate copyright notice and being sold through ordinary commercial channels.\n\nWhy would the market suddenly stop treating that book as proprietary?\n\nThe corporation will continue asserting ownership. Its distributors will continue selling it. Its licensees will continue paying for permission to use it. Competitors considering reproducing it will still face the possibility of a copyright lawsuit.\n\nNothing in the Copyright Office's position requires the corporation to establish its entire creative history before it can make those claims.\n\nThat distinction between the legal validity of a right and its ordinary commercial recognition is central to understanding what will happen next.\n\n## What happens when someone actually challenges ownership?\n\nImagine two companies, A and B.\n\nCompany B sells a successful illustrated product and claims copyright. Company A suspects that much of the artwork was generated by AI with insufficient human involvement to qualify for protection.\n\nCompany A wants to reproduce the artwork without paying.\n\nHow does it proceed?\n\nIt could simply reproduce the material and defend itself if sued. Alternatively, where there is a sufficient concrete legal controversy, it could seek declaratory relief.\n\nEither approach involves legal uncertainty, expense and risk.\n\nAnd the US legal system already has established procedures for resolving disputed copyright claims.\n\nUnder [17 USC §410(c)](https://www.law.cornell.edu/uscode/text/17/410), copyright registration made before or within five years after first publication provides *prima facie* evidence of the copyright's validity and the facts stated in the registration certificate.\n\nThis is a rebuttable presumption, not an absolute guarantee. A credible challenge can overcome it. But it means a registration isn't simply an unsupported assertion that must be proved afresh whenever someone questions it.\n\nThe Copyright Office also requires applicants to [disclose more-than-trivial AI-generated content](https://www.copyright.gov/ai/ai_policy_guidance.pdf) and exclude unprotectable material from the scope of their registration claims. Inaccurate applications can create serious problems for claimants.\n\nAnd registration is no longer optional for a US work whose owner wants to bring an ordinary federal infringement action: [*Fourth Estate v. Wall-Street.com*](https://www.law.cornell.edu/supremecourt/text/17-571) confirmed that registration, or refusal of registration, generally must precede such a lawsuit.\n\nThese requirements create a genuine incentive for businesses to document human authorship, particularly for commercially important works.\n\nBut notice the scale at which they operate.\n\nA copyright claimant doesn't generally need to prove the creative history of every work before publishing it. Nor must every potential rights holder publicly establish that a work contains no AI-generated expression.\n\nIf a dispute goes to court, the claimant may need substantial evidence of human authorship. Where a registration presumption has been effectively rebutted, or no applicable presumption exists, uncertainty may ultimately defeat the claimant's case. The copyright owner still has to establish a protectable right.\n\nThat is a real risk.\n\nIt is not, however, a new principle that uncertainty automatically places the disputed material in the public domain.\n\n**In the ordinary marketplace, the burden of deciding whether to challenge an asserted copyright remains with whoever wants to act contrary to that claim.**\n\nAnd that practical burden matters enormously.\n\n## The value of copyright is determined in the market\n\nThe more interesting question is not whether a particular work is legally copyrightable.\n\nIt is whether doubts about copyrightability materially reduce the economic value of a corporation's intellectual property.\n\nFor companies whose value depends heavily on copyright portfolios, intellectual property is a commercial asset. Rights generate income through licensing, distribution agreements, publishing, merchandising and countless other arrangements.\n\nInvestors value that property because they expect it to produce future cash flows.\n\nThey don't require absolute certainty that every underlying right would survive every conceivable legal challenge.\n\nThey require commercially sufficient confidence.\n\nThis is already true of ordinary intellectual property. Patents can be invalidated. Copyright ownership can be disputed. Licenses can contain defects. Contracts can be challenged. The commercial value of those assets nevertheless persists.\n\nA useful way to think about the question is that the economic impact of an authorship defect depends on the probability of a successful challenge, the consequences if one succeeds, and the costs and disruption associated with defending it.\n\nIf those risks are material, they can reduce the price a buyer is willing to pay.\n\nIf they are remote, commercially tolerable or inexpensive to manage, they may have little effect.\n\nConsider a fictional entertainment company with a valuable character franchise. The franchise generates revenue from films, merchandise, games and licensing agreements.\n\nSomeone alleges that AI was involved in producing some recent artwork.\n\nDoes this make the entire franchise substantially less valuable?\n\nProbably not.\n\nThere may be existing human-authored characters, scripts, designs, trademarks, contracts and other rights supporting the business. AI authorship uncertainty concerning particular images need not threaten all of those interests.\n\nAt the other extreme, imagine a company whose principal asset is a catalogue of unmodified AI-generated illustrations, marketed on the promise that customers are buying exclusive rights to the images.\n\nThat company has a much more serious problem. If the images aren't copyrightable, competitors may be free to reproduce them, subject to any other applicable rights.\n\nThe difference is not whether AI was used.\n\nIt is how much of the business's value depends on the particular exclusivity that AI authorship might undermine.\n\nThis suggests a more selective commercial response than Doctorow's prediction.\n\nCompanies will devote resources to proving copyrightability where the value at risk justifies it.\n\nEverywhere else, they may simply continue operating as before.\n\n## Are companies actually introducing AI formalities?\n\nHere Doctorow has stronger evidence on his side than might initially appear.\n\nThere are already concrete examples of businesses introducing AI-specific documentation, warranties and restrictions.\n\nThe most revealing are not announcements about ethical AI or voluntary transparency. They are actual commercial agreements.\n\nIn April 2024, the National Venture Capital Association updated its model financing documents to include representations concerning generative AI use. Among other things, companies may be asked to confirm that their use of generative AI has not materially compromised intellectual property they intend to maintain as proprietary.\n\nLaw firm Mayer Brown examined these provisions in its [May 2024 analysis of AI representations in technology acquisitions](https://www.mayerbrown.com/en/insights/publications/2024/05/ai-specific-representations-in-tech-m-and-a).\n\nIts observations are particularly relevant: many AI-specific representations substantially duplicate protections already provided by ordinary intellectual property warranties. The firms are largely adapting existing transaction procedures to address newly identified risks.\n\nMore significantly, there are now examples in actual SEC-filed transaction documents.\n\nA [November 2025 merger agreement between Kratos Defense & Security Solutions and Orbit Technologies](https://www.sec.gov/Archives/edgar/data/1069258/000106925825000062/ex21agreementandplanofmerg.htm) contains representations addressing whether generative AI was used in ways believed to materially affect proprietary intellectual property rights.\n\nA [July 22, 2026 asset purchase agreement between Progress Software and Domo](https://www.sec.gov/Archives/edgar/data/876167/000155278126000390/e26303_ex2-1.htm) includes assurances regarding AI use in developing material proprietary IP, alongside commitments concerning AI controls and procedures.\n\nAnd an [October 27, 2026 merger agreement between EnerSys and Ultralife](https://www.sec.gov/Archives/edgar/data/875657/000143774926033731/ex_1032339.htm) goes considerably further.\n\nIts intellectual property provisions refer to a disclosure letter listing AI models used by the company, the purposes for which they were used, descriptions of material outputs, and how those outputs were employed.\n\nIt also contains representations addressing AI-generated outputs whose ownership or eligibility for intellectual property protection may be uncertain.\n\nThis is real evidence of the private sector creating additional administrative requirements around AI.\n\nBut it is also remarkably revealing about what form those requirements are taking.\n\nThe language is concerned with *material* outputs, proprietary intellectual property, commercial risk, and the allocation of responsibility between buyer and seller.\n\nIt is not a universal public system for documenting the authorship of every piece of expression.\n\nNor does the existence of contractual representations demonstrate that a business has implemented exhaustive provenance tracking. The disclosure records are generally private, and the degree of scrutiny behind particular assurances is not visible from the public agreements alone.\n\nWhat we can observe is a process of private due diligence becoming more explicit about AI.\n\nThat is a significant development.\n\nBut it is recognisably an extension of existing commercial practice.\n\n## Different industries are responding differently\n\nThe broader evidence is also instructive.\n\nAdobe Stock [accepts generative AI images, vectors and videos](https://helpx.adobe.com/stock/contributor/submit-your-content/submit-generative-ai-content/submit-generative-ai-content.html), provided contributors comply with its licensing and disclosure requirements. Contributors must identify qualifying AI-generated submissions.\n\nThis is a formality, but it is not a requirement to prove that every image is entirely human-authored.\n\nIt is a classification and risk-management mechanism operating inside an existing licensing business.\n\nGetty Images, by contrast, [does not accept AI-generated or AI-modified editorial submissions](https://www.gettyimages.com.au/workwithus/editorial).\n\nThat makes commercial sense for a different reason. Editorial photography is valuable partly because customers expect it to document actual events. Authenticity matters independently of copyrightability.\n\nShutterstock takes another approach. Its [generative AI licensing arrangements](https://www.shutterstock.com/help/en/articles/11962590-are-my-images-indemnified) can include contractual indemnification for eligible images that have passed human review.\n\nHere the seller is managing legal risk and providing commercial reassurance. Indemnification doesn't magically create copyright in otherwise uncopyrightable expression, but it can make the transaction more attractive to customers.\n\nIn entertainment, Netflix has [published guidelines for production partners using generative AI](https://www.screendaily.com/news/netflix-releases-guidelines-for-using-generative-ai-in-productions/5208200.article). Certain uses require disclosure or approval, particularly when generated material may appear in final productions or involve major creative elements, performers or other rights holders.\n\nMeanwhile, the Writers Guild of America's [2023 agreement](https://www.wga.org/contracts/know-your-rights/artificial-intelligence) established requirements concerning AI-generated material supplied to writers. Its [2026 agreement](https://www.wga.org/contracts/contracts/mba/summary-of-the-2026-wga-mba) added notification requirements where companies license writers' work for training commercial generative AI systems.\n\nThese are meaningful contractual controls, some of which directly advance the interests of human creators.\n\nAnd there are technical formalities emerging too. Adobe [automatically attaches Content Credentials to certain Firefly-generated assets](https://helpx.adobe.com/firefly/web/get-started/learn-the-basics/content-credentials-overview.html), using provenance metadata to disclose information about how content was produced.\n\nBut Content Credentials primarily establish and communicate a record of provenance. They are not a judicial determination of copyright validity, and they don't create ownership rights merely by recording an assertion.\n\nTaken together, these developments show that AI is changing commercial procedures.\n\nThey do not establish that companies are being forced to prove human authorship for every expressive element in everything they sell.\n\nIndeed, the diversity of responses suggests something else: **markets are finding different ways to accommodate different kinds of AI-related risk.**\n\nSometimes that means disclosure. Sometimes it means excluding AI-generated content. Sometimes it means licensing, indemnification or contractual assurances. Sometimes it means making AI use an ordinary part of the product.\n\nThere is no single inevitable response.\n\n## Why these aren't the formalities Doctorow is imagining\n\nDoctorow's larger historical argument concerns the disappearance of copyright formalities.\n\nHistorically, the United States imposed administrative requirements that could determine whether copyright protection was obtained or maintained. These included notice, registration, deposit and, under earlier regimes, renewal.\n\nThe modern system differs fundamentally.\n\nCopyright generally arises automatically when an original work of human authorship is fixed in a tangible medium, under [17 USC §102](https://www.law.cornell.edu/uscode/text/17/102).\n\nRegistration remains important for enforcement and certain remedies, but it is not ordinarily a condition for copyright to exist.\n\nThe United States implemented the Berne Convention through legislation enacted in 1988 and joined the Convention effective March 1, 1989—not in 1976, as Doctorow's article suggests. The [Copyright Office confirms that date](https://copyright.gov/help/faq/definitions.html).\n\nNone of the corporate practices described above changes the fundamental operation of that system.\n\nA private buyer can require detailed representations before acquiring a company. A stock photography marketplace can demand labels. A film studio can impose production rules. A platform can require metadata.\n\nBut none of those procedures becomes a general condition for enjoying copyright protection under American law.\n\nAnd the incentives driving them are quite different from those of a public copyright registry.\n\nA public formality can make rights easier for everyone to identify and challenge.\n\nA private disclosure process primarily protects the commercial interests of the parties involved.\n\nThe EnerSys–Ultralife agreement illustrates this distinction neatly. Its AI disclosure requirements are extensive, but the resulting information is supplied through a private transaction process.\n\nThe existence of those records doesn't mean the public gains access to them.\n\nCorporations can become better at documenting the property they claim to own without making it materially easier for outsiders to challenge those claims.\n\nThat may be the opposite of the public-interest result Doctorow hopes to see.\n\n## The deterrent effect of an uncertain copyright claim\n\nThere is another important asymmetry here.\n\nA corporation asserting copyright in a commercially valuable work can be quite comfortable with uncertainty about the ultimate outcome of a hypothetical lawsuit.\n\nIts competitors may be considerably less comfortable.\n\nSuppose a publisher suspects that a rival's bestselling illustration was generated entirely by AI.\n\nIt might be legally entitled to reproduce the image.\n\nBut how confident must it be before doing so?\n\nIf the rival has registered a copyright claim, maintains that substantial human authorship was involved, and has the resources to litigate, the decision is not straightforward.\n\nEven successful litigation can be expensive and disruptive.\n\nThe larger publisher may have a diversified portfolio, legal counsel and substantial cash reserves. A smaller competitor may have much more to lose from a single dispute.\n\nThis is not a guarantee that the larger publisher wins. A credible challenge can succeed, and litigation risks cut both ways.\n\nBut it creates a possible gap between *de jure* exclusivity and *de facto* exclusivity.\n\nA work can be treated commercially as exclusive even if some of its claimed copyright protection would fail under sufficiently close examination.\n\nThat doesn't make the claim legally valid.\n\nIt can nevertheless make the claimed property commercially valuable.\n\nThis phenomenon isn't unique to AI. It arises whenever enforcement costs, uncertainty and unequal resources influence how people respond to asserted legal rights.\n\nAI may create more opportunities to challenge copyright validity without necessarily making those challenges cheap enough to change ordinary market behaviour.\n\nIndeed, widespread uncertainty could favour established rights holders, who are better positioned to manage the disputes.\n\n## Does the threat of AI reduce the value of corporate intellectual property?\n\nSometimes, certainly.\n\nIf a company derives most of its economic value from exclusive rights over material that turns out to be uncopyrightable, the consequences could be severe.\n\nA potential buyer would be rational to investigate the issue. A lender might discount assets that cannot reliably support expected revenues. A licensee might demand additional assurances.\n\nThe SEC-filed agreements discussed above demonstrate that these questions are already entering transaction documentation.\n\nSo Doctorow is right that provenance may become increasingly important.\n\nBut his prediction becomes much less convincing when generalised to the whole creative economy.\n\nA corporation does not need every creative element of every product to be independently copyrightable for its business to possess valuable intellectual property.\n\nHuman-authored material can remain protected within mixed works. Existing catalogues can retain their rights. Trademarks, contracts, exclusive access arrangements and other commercial assets can preserve value independently of copyright in particular outputs.\n\nThe potential loss of one category of rights doesn't automatically destroy the economic usefulness of the larger product.\n\nEven contractual terms from AI providers illustrate the difference between commercial ownership arrangements and the existence of statutory copyright. [OpenAI's terms](https://openai.com/policies/row-terms-of-use/) assign users whatever rights OpenAI may have in their outputs, while acknowledging that outputs may not be unique. Such terms help allocate rights between contracting parties. They cannot manufacture a copyright where the law recognises none.\n\nIn other words, companies can continue trading, licensing, valuing and making representations about AI-assisted products even when the precise extent of their statutory copyright protection is uncertain.\n\nThe market can live with that ambiguity.\n\nIt already lives with plenty of others.\n\n## The greater threat may be competition, not copyrightability\n\nThere is a final distinction worth making.\n\nAI may substantially undermine the economic value of some intellectual property businesses even if their copyright claims remain completely defensible.\n\nConsider the stock imagery market.\n\nA company with a large catalogue of human-created photographs might retain perfectly valid copyrights in every image it owns.\n\nBut if customers can generate acceptable substitutes at negligible cost, those copyrights become less commercially valuable.\n\nNot because they have been invalidated.\n\nBecause customers no longer need the protected works as much as they once did.\n\nThis is a much more direct economic mechanism.\n\nCopyright prevents certain unauthorised uses of protected expression. It does not prevent others from producing different works that satisfy the same customer demand.\n\nGenerative AI may make that competition dramatically easier.\n\nThe price of exclusive rights to a particular image can fall even when those exclusive rights are legally impeccable.\n\nThe same argument potentially applies to commercial illustration, advertising copy, routine software development, music and other markets where AI lowers the cost of producing substitutes.\n\nIf corporations find their creative assets losing value, this may be a far more important cause than uncertainty over who authored particular outputs.\n\nAnd unlike a legal challenge that might never happen, competition can affect revenue immediately.\n\n## What I think will actually happen\n\nThe evidence points towards a fairly conventional commercial adaptation.\n\nCompanies will continue claiming copyright over their products. They will continue publishing copyright notices, registering commercially important works, licensing their catalogues and asserting ownership in transactions.\n\nBuyers and investors will increasingly ask whether AI use has materially compromised the rights being sold.\n\nContracts will acquire additional warranties. Some industries will introduce disclosure requirements. High-value productions will maintain more detailed creative records. Platforms will develop provenance metadata.\n\nWhere the risk is substantial, these requirements may become rigorous.\n\nWhere it is insignificant, they may remain minimal or disappear into ordinary business procedures.\n\nThis is not a prediction that AI provenance won't matter. It clearly already does.\n\nIt is a prediction that companies will respond to it in proportion to commercial incentives rather than adopting a universal new standard of copyright documentation.\n\nCrucially, the evidence so far supports the emergence of targeted contractual controls, not an industry-wide requirement to track the human or machine origin of every expressive element.\n\nThose are very different outcomes.\n\n## Formalities for whom?\n\nDoctorow sees a possibility that AI will force corporate copyright holders to treat creative rights with more seriousness and, perhaps, create a system that works better for artists and the public.\n\nSome of the changes he describes may indeed benefit creators. Better records of human contributions could make attribution and compensation easier. Stronger agreements could prevent employers from using AI to evade obligations to writers and artists.\n\nBut there is nothing inevitable about that distribution of benefits.\n\nA corporation can document its provenance privately, satisfy its financiers, maintain its proprietary claims and give the public almost no additional information.\n\nIt can insist that contractors disclose their AI use without making equivalent disclosures about its own productions.\n\nIt can demand warranties from smaller suppliers while preserving ambiguity in its dealings with competitors.\n\nIt can afford the legal advice, compliance systems and litigation required to navigate uncertain copyright claims.\n\nSmaller businesses and individual creators may be less able to do so.\n\nIf private copyright formalities become widespread, they may strengthen the ability of large corporations to assert and defend intellectual property rather than weakening it.\n\n**The real question is not whether companies will be required to do more paperwork. It is who gets access to that paperwork, who bears its cost, and who gains bargaining power as a result.**\n\nDoctorow is right to identify a tension between generative AI and the doctrine of human authorship.\n\nHe is also right that we are seeing the emergence of new private procedures for managing that tension.\n\nBut there is a missing step in his argument.\n\nWhy should those developments cause the market to stop accepting corporate claims of ownership?\n\nWhy should they fundamentally diminish the commercial value of those claims, except where particular risks become demonstrably material?\n\nAnd why should the resulting private systems serve the public interest rather than the interests of the companies implementing them?\n\nThe more likely outcome is not the restoration of copyright formalities as a public institution.\n\nIt is the expansion of ordinary corporate due diligence, contractual risk allocation and selective provenance management.\n\nBusinesses will still claim ownership. Markets will still assign value to those claims. Competitors will still weigh the risks of challenging them.\n\nThe legal limits of AI authorship will matter whenever they become commercially consequential.\n\nBut until then, copyright will continue to operate much as it does today: partly as a system of enforceable statutory rights, and partly as a market of claims that most participants find it cheaper to respect than to contest.\n\n**AI may change what corporations can legally own. That doesn't mean it will quickly change what they claim to own—or what the market is prepared to pay for it.**\n\n\n**Australian companion:** [No Register, Plenty of Copyright Claims](/topics/ai-and-society/australian-copyright-claims) examines the equivalent questions under Australian law, including its lack of copyright registration and its rules for disputed authorship.\n\n---\n\n*AI disclosure: Generative AI was used in the production of this article. How much? I'm not telling. ;-)*","hero":"01a19af0-8e61-70a9-b16a-d095173a0405","media":[{"asset_id":"01a19af0-8e61-70a9-b16a-d095173a0405","alt":"Illustration of an asserted copyright certificate, courthouse and price tag"}],"author":"Emlyn O'Regan"}}