The governing copyright text is found in Title 17 of the United States Code (USC) and includes the Copyright Act of 1976, and all amendments (the “Copyright Act”). The language of the Copyright Act can be located at copyright.gov, the official website for the US Copyright Office (CO).
The CO regulations regarding registration practices are codified and published in the Code of Federal Regulations (CFR) under Title 37 — Patents, Trademarks, and Copyrights, Section II.
The CO has also released revisions to the US Copyright Office, Compendium of US Copyright Office Practices (3rd edition, 2017), the governing administrative manual for registrations and recordations issued by the CO.
The Copyright Act is interpreted by US federal courts.
The USA became a party to the Berne Convention Treaty on 1 March 1989 and has adhered to the World Intellectual Property Organization (WIPO) Copyright Treaty since 6 March 2002 and the WIPO Performances and Phonograms Treaty since 20 May 2002.
The USA is a member of the World Trade Organization (WTO) and follows the Agreement on Trade-Related Aspects of Intellectual Property Rights (TRIPS), which sets forth obligations related to intellectual property rights, including copyright and enforcement.
The USA is also party to other significant international conventions and treaties, including:
The Berne Convention permits foreign copyright holders to bring infringement claims within the USA even when the work does not have US copyright registration. Section 104 of the Copyright Act pertains to the national origin of works receiving US copyright protection and copyright in restored works.
This being said, registration is a prerequisite to certain remedies, specifically regarding statutory damages or attorney’s fees.
To qualify for copyright protection, Section 102 of the Copyright Act requires that a work be an “original work of authorship” and “fixed in any tangible medium of expression”.
The first element does not present a particularly high bar but works must display at least a minimal level of creativity, which tends to raise questions over copywriting works of conceptual art or new technologies. The second element requires a work to be captured and perceived, reproduced, or communicated for more than a short time.
Copyright protection is automatic from the moment of creation, but authors can receive enhanced benefits from registration.
Copyright notice has not been required upon publication of a work since 1 March 1989 when the USA joined the Berne Convention treaty, however, owners preserve certain litigation benefits by publishing a work with proper notice.
Proper notice consists of the copyright symbol © or the word “Copyright”, the name of the copyright owner, and the year of first publication. Registration establishes a claim to copyright with the CO. Although registering with the CO is not mandatory, for US works, it is a prerequisite to filing a claim in federal court to enforce one’s copyright and it permits authors to seek remedies under the Copyright Act in litigation. The CO serves as the register of copyright, and applications for certificates of registration can be filed online (or on paper).
Registrations may be searched for in the Records Room of the CO or a request may be made for the CO to conduct a search (for a fee of USD200 per hour or a fraction thereof, with a two-hour minimum). The printed Catalog of Copyright Entries covers registrations from 1891 through to 1978. An electronic version is available for works registered from 1978 to the present. To individually search public records, one can visit the Public Records Room at the Library of Congress. Where records are unavailable for public search, the CO will conduct the search, for a fee.
Consulting informational circulars published by the CO is recommended before conducting a search.
Section 102 of the Copyright Act provides categories of works protectable by copyright, including:
These categories should be interpreted broadly; for example, computer software is protected under “literary works”, while maps and technical drawings fall under “pictorial, graphic, and sculptural works”.
Software is protected as a literary work, receiving the same protections. See the CO’s Circular 61, containing helpful information regarding the registration of source codes.
Each version of a published software program containing new, copyrightable authorship requires a new application. The subsequent registration will not cover a previously published or registered source code, a source code in the public domain, or a copyrightable source code owned by a third party.
The deposit requirement for software permits filing the entire source code or a representative sample, which is useful if the source code contains trade secrets.
For the purposes of registration, hypertext markup language (HTML) is neither considered a computer program nor software. The CO permits registering HTML as a literary work if created by a human rather than a website design program, provided it contains sufficient creative expression. To register HTML, a complete copy of the code must be submitted. See Circular 66 on “Websites and Website Content”.
Databases are not a protected category of creative works, however, a database comprised of a collection and arrangement of material may be registered as a compilation of copyrightable work. This form of authorship recognises the selection, co-ordination and arrangement involved in creating a database. Additionally, material appearing in a database may be deemed authored and thus protectable.
Either form may be registered with the CO, provided that the database contains sufficient original expression and the copyright claimant owns the copyright in that material.
The CO maintains specific regulations for the registration of a photograph database.
Industrial designs do not receive copyright protection but may qualify for a design patent under 35 USC Section 171, which protects the ornamental design of an article of manufacture.
“Author” is undefined in the Copyright Act. Copyright initially vests in the individual who created the work, unless the work qualifies as a “work made for hire”, either because the individual is an employee and created the work within the scope of their employment or because the work falls within the categories of works that can be created as a work for hire. Entities can own work either because it is created as a work for hire or because the copyright is assigned to the entity. The CO will only register works created by a human.
Copyright can be jointly owned provided the work was prepared by two or more authors who both intended that their contributions combine into “inseparable or interdependent parts of a unitary whole”. Joint authorship arises either where authors directly collaborated or where they prepared their contributions separately but knew and intended that the contributions would merge. Paramount is the intention, at the time of creation, for the contributions to form an integrated unit. A contribution to a joint work is “inseparable” if the work contains a single form of authorship, as with a novel or painting, and “interdependent” if the work contains multiple forms of authorship, such as a motion picture or the music and lyrics of a song.
A copyright applicant must determine whether a work qualifies as a joint work based on the facts surrounding its creation. This determination bears importantly on the ownership and term of the copyright. For example, the authors of a joint work jointly own the copyright in each other’s contributions, as well as an undivided interest in the copyright for the work as a whole. In other words, each enjoys an independent right to use or license a joint work, provided they account for the profits. A joint author may exploit the joint work but may not grant exclusive rights, or transfer all interest in the work, without obtaining consent from the other author(s). If a joint author does exploit a work non-exclusively, they must account to the other author(s).
The copyright term for a joint work expires 70 years after the death of the last surviving author. If the joint work was created by two or more authors as a work made for hire, an anonymous work, or a pseudonymous work, the copyright expires 95 years from the year of publication or 120 years from the year of creation (whichever is shorter).
Authors may claim copyright anonymously or pseudonymously. In this case, the copyright lasts 95 years from the year of publication or 120 years from the year of creation (whichever is shorter). If, however, an author reveals their identity, this term converts to the life of the author plus 70 years.
The Copyright Act protects collective works, including anthologies or encyclopaedias, in which a number of contributions, constituting separate and independent works, are collectively assembled. The Copyright Act also protects compilations, including collective works comprised of pre-existing material arranged in a way that results in an original work of authorship.
Copyright in compilations or collective works only extends to the material contributed by the author of the collection, not to pre-existing material. Copyright in a collection does not enlarge the scope of copyright in the pre-existing material. For example, a collection that includes public domain material will not cause the public domain work to be protected by copyright.
The copyright in a collective work is distinct from the copyright in component works and can be exploited separately and apart from the pre-existing material. The owner of a collective work may only exploit the component parts as part of the collective work or any revision.
Any entity, including a corporation, may qualify as an author of a work in the USA under Section 201(b)’s “work made for hire” doctrine. Under the doctrine, an employer is considered the author unless the parties sign a written agreement to the contrary. Work made for hire applies under two scenarios: where the work is prepared by an employee within the course and scope of employment, and where the work has been specifically ordered or commissioned for use in nine specific categories:
For the second scenario, the parties must expressly agree in a written, signed instrument that the works shall be considered for hire. Apart from conforming with general contractual standards, there are no additional standards. For the first scenario, courts distinguish between work prepared by an independent contractor versus an employee, employing a multi-factor agency law analysis to examine the relationship, such as the hiring party’s right to control and the hired party’s discretion to perform the work. Alternatively, collective bargaining agreements in various industries designate a worker’s employee status and are often given weight by the courts.
Public employers operate under the same work-made-for-hire doctrine as other employers, although many universities have specific contracts with professors that address whether a professor’s writing is outside the scope of their employment and whether the professor would retain the copyright.
The Copyright Act does not accord protection for any work of the US government; however, the government may still hold copyrights transferred by an assignment, bequest, or otherwise.
Section 106 of the Copyright Act accords six exclusive economic rights to owners:
These rights persist throughout the term of copyright and do not affect a term’s duration. For works created on or after 1 January 1978, copyright persists for the life of the author plus 70 years, or, in the case of jointly authored works, for 70 years after the last surviving author’s death. For works made for hire and anonymous or pseudonymous works, copyright lasts 95 years from first publication or 120 years from creation (whichever is shorter), although in the latter category, if the author’s identity is revealed, the term becomes the author’s life plus 70 years. Copyright terms do not otherwise vary based on the type of work.
Works created and published prior to 1978 were initially protected for 28 years plus a 28-year renewal term, but this term has since been extended many times over by amendments, and pre-1978 unpublished works are protected by the Copyright Act. For additional information, see Circular 15A “Duration of Copyright”.
Transferred Rights and Termination
An author or their surviving spouse or heirs may recapture transferred rights for works not made for hire under Sections 203 and 304 of the Copyright Act. Section 304 applies to transfers and licences executed before 1 January 1978, for works in their first or renewal term by that date. Section 203 relates to transfers and licences executed on or after 1 January 1978.
Grants may only be terminated during a specific statutory window and notice must be served on the original grantee(s) within a certain time period, in order to be effective. Terminations under Section 203 may generally be effectuated during a five-year window from the 35th anniversary of the grant, while terminations under Section 304 may be effectuated during a five-year window from the 56th anniversary of the date on which copyright was secured.
Termination aims to afford authors and successors a “second bite of the apple”, allowing them to regain rights they may have granted when they lacked bargaining power. However, under this system, only US rights revert; any foreign granted rights will not return to an author or successors upon termination. Furthermore, derivative works previously created under a grant may continue to be exploited following termination. However, when an author dies during the first term of a work’s copyright (for pre-1978 works), case law has established that their heirs will not be bound to any agreements the author made permitting use of derivative works during the renewal term.
Additional information may be found on the Copyright Office’s website.
A copyright owner may transfer their exclusive rights within a copyright. A transfer, other than by operation of law, is only valid if in writing and signed by the owner of the rights being transferred, or by the owner’s representative.
A transfer is effectuated by an assignment, mortgage, exclusive licence, or other conveyance and may include the copyright in its entirety or limited exclusive rights. Limitations on the time, place, or effect of a transfer will not affect its validity.
A certificate of acknowledgement is not required for a valid transfer but is considered prima facie evidence if executed by a person who administers oaths within the USA, or, in the case of a transfer executed in a foreign country, if issued by a diplomatic or consular office of the USA or a person authorised to administer oaths. Transfers may be recorded with the CO but this is not required.
Economic rights are transmissible upon death. Section 201(d) of the Copyright Act provides that copyright ownership may be transferred under the law by will or as personal property, in whole or in part.
Copyright can be transferred. Any transfer of exclusive rights must be in writing and signed by the owner or representative. Minors (in the USA, those under 18 years of age) may claim and exercise copyrights, but state law may regulate transfer agreements entered into by minors.
Section 109 of the Copyright Act recognises the first sale doctrine (commonly called an “exhaustion” doctrine outside the USA), by which the owner of a work’s copyright may not prevent subsequent sales of the work where there has been a first, authorised sale of a copy. Essentially, the owner’s right to control distribution of a lawful copy is “exhausted” following the first authorised distribution. An actual “sale” is not a necessary condition for application of the doctrine but, at the very least, a copy must have been lawfully made and conveyed. Digital content, like an e-book, is not included in this doctrine because it is leased rather than owned.
The USA recognises moral rights within visual arts, defined within Sections 106A and 101 of the Copyright Act as paintings, drawings, prints, or sculptures existing in a limited edition of 200 copies or fewer that are signed and consecutively numbered by the author (and bearing their signature or other identifying mark), as well as still photographic images produced for exhibition purposes.
Moral rights include the rights of attribution and integrity. The right to attribution allows an author to be credited as the author of their work, while the right of integrity allows an author to prevent intentional and prejudicial distortions of the work, as well as the destruction of a work of “recognised stature”. Moral rights also allow authors to prevent others from using their name in association with a work they did not create.
Moral rights only apply to a work’s author and persist only throughout the author’s life, or, in the case of joint authorship, throughout the life of the last surviving author. Their term runs to the end of the calendar year in which they would expire. Moral rights may not be transferred but may be waived if an author expressly agrees to a waiver in a signed, written instrument identifying the work and the uses of the work to which the waiver will apply.
Moral rights are distinct from ownership of a copy of any work. Consequently, any transfer of ownership of a copy of a work of visual art or any rights under copyright does not, without the written, signed instrument discussed above, constitute a waiver of an author’s moral rights.
The USA has implemented anti-circumvention measures along the lines of Article 11 of the WIPO Copyright Treaty by way of the Digital Millennium Copyright Act of 1998 (DMCA), codified in Section 1201 of the Copyright Act.
Section 1201(a)(1) prohibits any person from circumventing a technological measure that effectively controls access to a work protected under the Copyright Act. Subsection 1201(a)(3) defines this circumvention as a “means to descramble a scrambled work, to decrypt an encrypted work, or otherwise to avoid, bypass, remove, deactivate, or impair a technological measure, without the authority of the copyright owner”. For example, bypassing a paywall to read a news article without paying for a subscription would fall under Section 1201(a)(1). Additionally, the law prohibits trafficking in technologies that circumvent technical measures.
The statute does provide certain exemptions and exceptions in Sections 1201(d)–(j), however, including:
Persons injured by a violation of Section 1201 may bring a civil action and seek the actual damages suffered, including profits of the violator attributable to the violation, or elect statutory damages for each violation ranging from USD200 to USD2,500.
The USA has implemented legal remedies with respect to copyright management information (CMI) along the lines of Article 12 of the WIPO Copyright Treaty by way of DMCA Section 1202.
Section 1202(a) prohibits any person from knowingly, and with the intent to induce, enable, facilitate, or conceal infringement, from either providing false CMI, or distributing or importing false CMI. Section 1202(b) also prohibits the intentional removal or alteration of CMI, as well as the distribution, importation, and public performance of works knowing that the CMI has been removed or altered. Section 1202(c) broadly defines CMI as including, among other things, title and copyright notice information, and identifying information about the author, owner, writer, or performer who is credited in an audio-visual work.
Persons injured by a Section 1202 violation may bring a civil action and seek the actual damages suffered, including profits of the violator attributable to the violation, or elect statutory damages for each violation, ranging from USD2,500 to USD25,000.
There are several collective rights management (CRM) systems in the USA that cater to different types of copyrighted works, major examples of which are referenced below. However, CRM is complicated and industry-specific, and while it is governed by statute to some degree, US antitrust laws and existing court consent decrees render participation in CRM organisations voluntary.
Music
US music copyrights are split into two parts: the underlying musical composition copyright (the written music and lyrics), and the sound recording embodying the composition’s performance. Different organisations collect and disburse earned royalties to the appropriate rightsholder.
With respect to composition copyrights, the songwriter(s) and music publisher(s) typically split the income earned from certain uses of the composition. For public performances (such as on radio, TV, and in public venues like concert halls, and even bars and restaurants), performing rights organisations (PROs) collect money from each entity holding a licence issued by the society for the use of musical works in the PRO’s repertory and distribute it to their members based on proprietary formulae. PROs also enforce the rights of their members against any unlicensed public performance of their music. There are various licences available but generally one pays an annual fee for a “blanket” licence that covers the entire repertory of the PRO. The two largest PROs are ASCAP and BMI, which comprise about 96% of the music market, followed by SESAC and newcomer GMR.
Composition copyright owners also employ agencies to collect “mechanical” royalties, or moneys earned by songwriters and publishers from the creation and sale of sound recording renditions of a composition. The Harry Fox Agency (HFA) traditionally issued these mechanical licences (“compulsory licences”) and collected mechanical royalties for its members. However, under the Music Modernization Act (MMA), statutory compulsory licences are being streamlined, replacing the old HFA song-by-song licensing structure with a blanket licensing system for digital music providers to make and distribute downloads and streams of digital music (HFA will still administer royalties for physical music sales such as CDs and vinyl). A new entity, the Mechanical Licensing Collective, started administering the blanket licences, receiving reports from digital music providers, and collecting and distributing royalties on 1 January 2021.
With respect to sound recording, a record label typically owns the copyright in the “master” of a sound recording. The label typically tracks record sales, whether physical or digital, while a collection society called SoundExchange tracks digital public performance royalties for sound recordings. Under the Copyright Act, sound recording copyright owners, unlike composition owners, lack an exclusive public performance right other than by digital means (ie, radio stations do not pay sound recording owners’ royalties for radio play).
Visual Arts
The Artists’ Rights Society (ARS) is a US licensing and monitoring organisation for visual artists, which represents the copyright interests of thousands of global visual artists (and estates), from painters and photographers to sculptors and architects. Among other things, ARS handles enquiries for licences to reproduce visual art works (eg, for publication in books, museum exhibitions and monographs). Membership is voluntary.
The American Society for Collective Rights Licensing is a non-profit corporation that collects and distributes funds earned abroad in connection with visual art to rightsholders in the USA, as well as to foreign rightsholders whose works are published in the USA. Membership is voluntary.
Publishing
The Copyright Clearance Center is a corporation providing collective copyright licensing services for users of copyrighted works in the corporate world and academia. It negotiates and secures agreements with copyright owners (particularly academic publishers) and arranges for collective licensing of the owners’ publications as the owners’ agent. Membership is voluntary.
See 6.1 Collective Rights Management System.
Using a copyrighted piece of music alongside visual or audio-visual material requires a synchronisation (or “sync”) licence.
Sync rights in the USA must be procured by the party seeking to use the music in a movie or TV show (usually the producer), and obtained from both the songwriter or music publisher and the sound recording owner. Sound recording sync rights are typically negotiated directly with the record label for the master rights, whereas publishing sync rights for the composition can be negotiated through HFA in many instances, or otherwise directly with the songwriter or publisher.
The USA employs a “fair use” system that permits use of copyrighted works without the copyright holder’s consent in certain circumstances. Codified in Section 107 of the Copyright Act, fair use may commonly be found in instances of “criticism, comment, news reporting, teaching (including multiple copies for classroom use), scholarship, or research...” although this list is not exhaustive.
In assessing fair use, courts consider four factors articulated in Section 107:
Courts weigh these factors case by case, because fair use determinations are particularly fact-specific. Even so, courts commonly accord the most weight to factor 1, and through developments in case law, have added a gloss to it which considers the “transformative” nature of the work. Factor 4 is the second most heavily weighted factor, with the other two factors rarely tipping the scales, with the exception of a recent Supreme Court fair use case – Google LLC v Oracle America, Inc – where the second factor, the nature of the work, weighed heavily in the court’s decision that the copying of Oracle’s popular Java “declaring code” by Google, in creating an API for its own smartphone, was fair use.
The fair use framework is famously complicated and, as noted, fact-specific. Case law provides valuable insight into how courts may undertake this analysis but it is far from consistent or clear, and often varies by jurisdiction.
Various provisions within the Copyright Act protect forms of private copying although there is no express exception for private copying.
Education and Computer Programs
Fair use protects copying for certain educational purposes, such as making copies for classroom use and may also protect making backup copies of electronic media, recording time-shifted televised broadcasts, and non-commercial archiving of personal properties, such as CDs and DVDs. The Audio Home Recording Act (Section 1001) protects home copying of pre-recorded music, such as music airing over the radio, for personal use.
Under Section 117 of the Copyright Act, computer program owners are also permitted to make an archival copy, but it must be destroyed if the owner ceases to have rightful possession of the original. An owner may also lawfully possess a copy of a computer program made as part of the activation of a machine that lawfully contains the program, or for maintenance or repair purposes, but the copy must only be used for those purposes and destroyed immediately following their completion.
Libraries and Archives
Section 108 protects libraries or archives engaged in private copying, provided these organisations “reproduce no more than one copy or phonorecord of a work”. This exception requires that the reproduction be made without the purpose of commercial advantage, the organisation’s collection be open to the public or available to more persons than merely the researchers affiliated with the organisation, and the reproduction include a notice of copyright or similar “legend” stating that the work is copyright-protected.
Section 108 further distinguishes between copying of published versus unpublished works. The law permits libraries or archives to duplicate an unpublished work from their collection solely for purposes of preservation or deposit for research use in another library or archives, provided that any digital copy is not further distributed or shared publicly outside the organisation. Libraries or archives may duplicate published works solely for replacement purposes to safeguard against damage, deterioration, loss, theft, or obsolescence of its storage format, if they ensure that the unused replacement cannot be obtained at a fair price and that digital copies are not made available outside of the organisation.
Furthermore, should the work be within the last 20 years of its copyright term, a library or archives, and some non-profit educational institutions, may under limited circumstances copy it for “preservation, scholarship, or research”.
Section 120 of the Copyright Act establishes an exception to copyright protection in regard to the pictorial representation of architectural works. Although architectural works may themselves merit protection, if such a work is embodied in a building “located in or ordinarily visible from a public place”, its copyright will not prevent persons from painting, photographing, or pictorially representing the work, including the distribution or public display of such representations.
The USA has established an exception to copyright infringement for activities carried out by intermediaries, such as internet service providers (ISPs). Pursuant to the DMCA Section 512(a)–(d), an ISP may qualify for one of four “safe harbours”, which limit their liability for infringement claims based on “transitory digital network communications”, “system caching”, “information residing on systems or networks at [the] direction of users” and “information location tools”.
To qualify, ISPs must meet certain threshold criteria. They must first qualify as an online service or network access provider or operator, which most do, and must also have adopted and reasonably implemented a repeat infringer policy that provides for terminating subscribers in appropriate circumstances. Additionally, they must accommodate standard technical measures used by copyright owners to identify or protect copyrighted works. To merit the safe harbour concerning user-generated content, ISPs must also have a registered agent with the CO.
Having met this threshold, an ISP must then satisfy the requirements of a particular safe harbour. For Section 512(c), a commonly litigated safe harbour protecting against infringement claims arising from an ISP’s storage of copyrighted user-generated or submitted content, an ISP must demonstrate lack of actual knowledge that the material is infringing, or that, upon obtaining knowledge, it acted expeditiously to remove or disable access; that it hasn’t received a financial benefit from the activity in cases where it had the right and ability to control the activity; and that upon notification of the infringement, it responded expeditiously to remove or disable access. The requirements for eligibility for the other safe harbours may be found in the statute.
Satire and parody fall under the umbrella of fair use, assuming they meet the applicable standards. Parody is a classic example of fair use, especially because the nature of parody is to lampoon a specific work. Satire, however, typically appears as criticism of society or larger abstract concepts and may be harder to prove as fair use if one is not critiquing a specific work.
Fair use is typically seen as a First Amendment exception to copyright protection. Fair use embodies the intersection of freedom of expression and the limited monopoly conferred by the Copyright Act. Fair use incorporates significant aspects of commentary, criticism and news reporting, which are all central First Amendment concepts. Indeed, fair use, while typically termed a “defence” to copyright infringement is actually not a defence at all, it is a permissive use of a copyrighted work that furthers the goals of the First Amendment.
The USA does not recognise the concept of neighbouring rights.
See 8.1 Neighbouring Rights.
See 8.1 Neighbouring Rights.
Infringement occurs when a party who is not the author, legal or beneficial owner, or licensee exercises one of the exclusive rights of copyright holders under Section 106 of the Copyright Act or any rights in Sections 106A or 107 to 122, without permission, creates a work that is substantially similar to an existing copyrighted work, or imports copies or phonorecords of a copyrighted work into the USA in violation of Section 602.
Direct and Secondary Infringement
Infringement can be direct or secondary. Direct infringement involves an accused infringer who undertook volitional conduct to commit the infringement, whereas secondary (indirect) infringement arises when a party contributes or induces a third party to commit the direct infringement or otherwise is vicariously liable for another’s direct infringement.
Contributory infringement occurs where the accused secondary infringer knew of the direct infringement and either caused, induced, or materially contributed to it. Vicarious infringement occurs where the accused secondary infringer had the right and ability to control or supervise the activity giving rise to the infringement and a direct financial interest in the same.
The many defences against a claim of copyright infringement, include the following.
Copyright holders may pursue infringement claims through federal lawsuits, as well as before the Copyright Claims Board (CCB), a tribunal within the Copyright Office established by the Copyright Alternative in Small-Claims Enforcement Act (the “CASE Act”) of 2020, which launched in June 2022 and will hear (and has started to hear) copyright infringement matters of lesser value.
Before filing a federal lawsuit, one must register the copyrighted work with the CO. Merely submitting an application is not enough; rather, the CO must have granted or denied the application (if denied, the applicant can still bring a claim but will not be entitled to a presumption of ownership or copyright validity). The CASE Act offers claimants expedited registration of works for a lower fee than otherwise charged by the Copyright Office. This approach is somewhat counterintuitive as work is considered “copyrighted” as soon as it is fixed in a tangible medium of expression, but it is necessary for purposes of enforcement. Bringing a claim before the CCB, however, only requires one to have submitted an application to register their work.
US federal courts have exclusive jurisdiction over copyright infringement actions. A federal statute, 28 USC Section 1338(a), states that copyright cases must be brought before a federal district court; this means that no state court may hear a copyright case. Appeals of copyright decisions filter up to the applicable Court of Appeals (divided geographically across the country).
The CASE Act
Pursuant to the CASE Act, the Copyright Office launched the CCB on 16 June 2022. The CCB provides a jurisdictional alternative to the federal court system and is a voluntary system for smaller-value copyright claims to be heard before a three-member tribunal presided over by qualified copyright attorneys. Litigants may seek rulings on infringement, declarations of non-infringement and certain DMCA claims; however, they may not seek injunctions. Parties do not require attorneys, personal appearances are unnecessary, and the process provides for only limited discovery.
The CASE Act caps statutory damages at USD15,000 per work with a case maximum of USD30,000, or statutory damages of up to USD7,500 per work not timely registered under Section 412 of the Copyright Act. The CASE Act also imposes sanctions for bringing frivolous or abusive claims, including barring litigants from bringing claims for 12 months and awarding attorney’s fees to the defendant(s) in bad-faith claims, up to USD5,000.
In its first six months of operation, the CCB has registered over 260 cases, primarily infringement claims, and largely without attorney representation. Additional education efforts are underway by the CCB and related parties, such as the US Copyright Office, as self-represented cases have, thus far, tended to warrant amendment or dismissal at greater rates than those involving attorneys; completing service of process and understanding the elements required to prove copyright infringement have been areas of challenge to unrepresented claimants. Considering the nascency of CCB operations, the creative community’s familiarity and understanding of CCP procedures is expected to improve, decreasing the number of cases requiring dismissal.
In the event that claims do not comply with CCB procedures, claimants are given 30 days to amend, with a second 30-day window for further amendment, as needed. Once the CCB approves claims for service of process, a claimant has 90 days to complete service, and respondents have 60 days after service to choose to participate in the proceeding or opt out. This option ensures that the CCB remains a voluntary process.
Foreign claimants may file claims before the CCB; however, neither foreign nor US residents may file claims against a foreign respondent, although a US resident who is a respondent may file counterclaims against a foreign claimant.
Further information about the CCB and how to bring and maintain a claim is located at https://ccb.gov/.
Only a copyright owner/author, assignee or exclusive licensee may bring an infringement action.
Third parties may be involved in proceedings for copyright infringement in several ways. When a plaintiff asserts claims of both direct and contributory infringement, and only pursues one type of offender, the other alleged infringers may remain at large but may end up as witnesses in the pending case.
Under the Federal Rule of Civil Procedure (FRCP) 65(d)(2), federal court injunctions bind the parties, the “parties’ officers, agents, servants, employees, and attorneys”, and “other persons who are in active concert or participation with” any of the above. Non-parties who do not receive notice of the injunction are not bound. Non-parties probably will not be sanctioned (unless they fail to comply with a non-party subpoena under FRCP 45) and will not be subject to damages awards.
In federal court, plaintiffs may seek emergency relief by filing for a temporary restraining order (TRO) or preliminary injunction (PI). These tools are governed by FRCP 65. TROs can be heard ex parte in court (without the participation of the party to be enjoined), are inherently limited in scope and time, and will only last for 14 days, whereas PIs necessitate full briefing and a hearing and, if granted, last throughout the litigation. The requirements are generally the same:
With TROs, movants can show, as an alternative to the likelihood of success on the merits, that sufficiently serious questions on the merits provide fair ground for litigation and must show that the balance of hardships tips decidedly in their favour, with irreparable harm imminent and likely.
Courts may, at any time during an infringement action, order infringing goods to be impounded.
Expert witnesses may be critical in certain copyright cases, but are not mandatory. In the context of substantial similarity, experts carry less weight than the “ordinary observer” test, which recognises a work’s “total concept and feel” (which should be decided by the trier of fact), but they can be useful where objective analysis is required.
Where appropriate, experts serve an important role in litigation, and qualified expert witnesses are used to submit a report and/or testify (or to simply consult) to support the litigant’s position, whether it be to help establish (or disprove) substantial similarity for an infringement analysis or support a damages theory. Experts are key in cases involving nuanced or technical areas of copyright law, including music and software.
Both the plaintiff and the defendant may engage experts who offer competing viewpoints, and use of a knowledgeable expert may spell the difference between a win and a loss. However, experts can be expensive, driving up the costs of litigation, leaving even successful litigants unable to recover the costs of their witnesses, especially considering the Supreme Court recently limited the types of costs awardable to prevailing parties under Section 505 of the Copyright Act.
US Customs and Border Patrol (CBP) officials have authority to seize potentially infringing goods. Seizures are governed by a procedure outlined in Title 19 of the Code of Federal Regulations. Copyright owners must first register their copyright with the CO and may then separately file an application to record the copyright to secure customs protection on the CBP’s Intellectual Property Rights e-Recordation website, or mail the application to a CBP office in Washington, DC, along with the USD190 fee. Once approved, CBP recordation lasts for 20 years, unless copyright ownership expires earlier. Owners may renew their recordation for USD80, as long as they submit the renewal application at least three months before the recordation expires.
If CBP officials see goods they suspect infringe a copyright recorded in the database, they may seize them at border control and withhold delivery. Specific notice is then given to both the importer and the copyright owner. The owner may engage in the administrative proceeding with the CBP or seek a court order enjoining importation of the article. Notably, even an owner who has not recorded their copyright with CBP may seek this court order. Following entry of the order, the owner must submit a certified copy of the same to the Commissioner of Customs in order to obtain CBP enforcement.
Monetary remedies for copyright infringement under Copyright Act Sections 504(a)–(d) include the copyright owner’s actual damages, including any of the infringer’s additional profits derived from the infringement, or statutory damages. Copyright owners must choose, prior to the rendering of a final judgment, whether to seek either actual damages and profits or statutory damages. There is a current split among US courts, including the Ninth and Second Circuit, over the amount of damages to award in cases where copyright holders first learn of infringement and bring suit long after it originally occurred – a practice allowed under the universally adopted “discovery rule”. Some courts will allow a copyright holder to recover damages dating back to the inception of the infringement, while others limit recovery to three years prior to the filing of the complaint.
Statutory Damages
Statutory damages range from a minimum of USD750 to a maximum of USD30,000 per work infringed (not per infringement) and do not require specific proof; in cases of wilful infringement, the courts have the discretion to increase a statutory damages award up to USD150,000. However, where an infringer can demonstrate that they were unaware of the infringing nature of their conduct (“innocent infringement”), the court may reduce the award to no less than USD200 per work infringed. A finding that the infringer reasonably believed the use to be fair use will protect limited classes of infringers against statutory damages, including employees or agents of non-profit educational institutions, libraries or archives.
Injunctive Relief and Equitable Remedies
The Copyright Act also provides for injunctive relief and equitable remedies. Under Section 502, district courts may grant temporary and final injunctions that are operative and enforceable nationwide against the targeted infringement. Section 503 allows courts to seize and impound infringing copies or phonorecords and, as part of a final judgment, to order the seized items destroyed.
The Copyright Act occasions criminal liability in limited circumstances. Under Section 506(a), criminal sanctions are available against a party that wilfully infringes a copyright for purposes of commercial advantage or private or financial gain, by reproducing or distributing one or more copies or phonorecords of a work valued at more than USD1,000 in any 180-day period, or by electronically distributing a work being prepared for commercial distribution with the knowledge that the work was intended for commercial distribution. Remedies for criminal infringement include forfeiture and destruction of infringing articles (Section 506(b)) and restitution, under 18 USC Section 2323 (a provision of the federal criminal code).
Some courts have held that a copyright registration is not required for the US government to bring a criminal complaint, but this is not settled law. An essential element of a criminal copyright infringement claim is the existence of a “valid” copyright, and a registration strongly evidences this.
Appellate procedures for copyright cases mirror any other federal case. Copyright cases are appealed to the US circuit courts of appeal, which are not specialised (with the exception of the US Court of Appeals for the Federal Circuit, which hears patent appeals, not copyright appeals). Appeals from the circuit courts proceed to the Supreme Court on a writ of certiorari, which the Court may or may not grant.
The basic US rule is that each party pays its own attorney’s fees and costs unless a contract or statute provides otherwise. Section 505 of the Copyright Act provides for the shifting of a prevailing party’s fees and costs to the non-prevailing party, but such awards are not mandatory, and district courts have discretion to determine whether fees should be awarded based on various factors established by case law. These include whether the non-prevailing party’s claims or defences were objectively unreasonable or frivolous, whether an award would further the prevailing party’s need for compensation and deter the non-prevailing party, and the non-prevailing party’s motivation.
Outside of the Copyright Act, other rules and statutes bear upon litigation costs, as copyright cases are, by nature, federal actions. Pursuant to FRCP 11, courts may impose sanctions, including paying another party’s attorney’s fees and expenses, on “any attorney, law firm, or party” found to have violated its responsibilities in conducting the lawsuit, including by bringing or maintaining a suit for an “improper purpose, such as to harass, cause unnecessary delay, or needlessly increase the cost of litigation”, as well as for bringing frivolous actions or making baseless factual contentions in court papers. FRCP 37(c), 28 USC Section 1927, and the courts’ inherent powers also allow for sanctions for various discovery and litigation abuses both against the parties and their attorneys.
Alternative dispute resolution (ADR) is a common way of settling copyright cases. While it is generally not mandatory, some courts strongly encourage (and may order) pre-trial ADR. The most common form in copyright cases is mediation, which is voluntary and non-binding, and can be held through a private provider such as the American Arbitration Association or JAMS, for which the participants must pay, or depending on the jurisdiction, may be provided for by the court at no cost or a reduced cost. Some courts appoint volunteer private attorneys as mediators at no cost; others provide a mediator for a certain number of hours at no cost with fees to be paid after a threshold is met; and others refer the case to a magistrate judge for mediation. Arbitration (which is binding and akin to a private trial) is uncommon in copyright cases and generally does not occur unless a contract between the litigants mandates arbitration versus litigation.
The Mediation Process
The parties select the forum and mediator, and if they cannot agree on a mediator, the organisation or court may appoint one. In copyright cases, parties tend to prefer mediators with substantive experience. Mediators typically hold an initial conference and request pre-mediation letters or briefs. The parties and their attorneys then convene an in-person mediation where the mediator takes turns speaking with each party until a deal is made or a stalemate occurs. Sometimes the mediator offers a “mediator’s proposal” that each side can either accept or reject confidentially.
41 Madison Avenue
Floor 38
New York
NY 10010
USA
+1 212 974 7474
+1 212 974 8474
nwolff@cdas.com www.cdas.comIntroduction
One of the hottest topics in copyright law in 2022 was generative artificial intelligence (AI). While AI has existed for nearly 50 years – and generative AI systems such as The Next Rembrandt have existed for several years – 2022 saw exponential growth both in the strength of the technology and the attention it has garnered across artistic industries and among intellectual property lawyers around the world.
Attracting the most discussion were text-to-image visual art creators including Craiyon, DALL-E 2, Midjourney, and Stable Diffusion (stability.ai), and photo editing and avatar-creation app Lensa. These platforms raised the buzz to a roar, allowing anyone to create artwork – from the complex and beautiful to the bizarre and disturbing – simply by typing a text “prompt”.
These generated images began going viral in April of 2022 and in months, OpenAI – the company behind DALL-E 2 – had over a million users, as did Midjourney’s pay service, Dream Studio. Text-to-video generators (such as Meta’s forthcoming Make-a-Video) are now on the horizon, as well as AI that can create all forms of copyrightable content including music, articles, poetry and code. These platforms are not without controversy and many have grabbed headlines over the past year, including an individual who used Midjourney to win a Colorado State Fair art prize over human-authored paintings. The public and commentators have mixed emotions over these new platforms given the complex legal and ethical gray areas generative AI raises. This article discusses some of the major copyright implications for generative AI systems under the US Copyright Act, including authorship, ownership, copyrightability, fair use and infringement.
What is Generative AI?
AI is any technique enabling computers to mimic human intelligence. Generative AI is governed by neural networks capable of “machine learning” or “deep learning” – processes that use information to solve new problems outside the scope of their initial training. In other words, generative AI includes unsupervised or semi-supervised algorithms that enable computers to use existing content like text, audio and video files, images, and even code to create new content that appears authentic. These computers detect patterns in the underlying datasets to output new content. The main generative AI models are Generative Adversarial Networks (GANs), which can create visual and multimedia content from both image and text datasets, and Generative Pre-Trained (GPT) language models that can use information gathered on the internet to create new content.
Text-to-image platforms can use a GAN or a diffusion model. The GAN model is based on one neural network trained to pair an image with text that describes that image, and another trained to generate new images that the first neural network accepts as a match for the prompt. Newer art generators such as DALL-E 2 use a diffusion model, a neural network trained to clean images up by changing the image pixel by pixel until nothing is left but static and then recreating the image, with each iterative version approaching a new image.
These processes occur using a massive underlying dataset from which the AI systems “learn” to create accurate content by matching text to images. These datasets, such as LAION, contain billions of pairings of text and images scraped from the internet and a tremendous amount of copyrighted works, creating difficult questions surrounding copyright infringement and fair use. But even before addressing dataset issues, fundamental questions at the heart of the AI debate are: (i) who can be an author of a generative AI work, and (ii) is the output from an AI engine resulting from the text prompt of a human user copyrightable?
Main Copyright Issues
Authorship and originality
A threshold question under US copyright law is who qualifies as the author of an AI-created work, assuming the generative output can be deemed original enough to be copyrightable. Most debates revolve around whether the AI itself can own the copyright, and if not, whether the user, programmer, or company that owns the AI can or should own the output – or if no one can or should own the resulting works. There are few clear answers, given the novelty of the legal questions.
AI as author
The clearest issue surrounding generative AI and copyright, is the unlikelihood that the AI alone will be deemed author of its output. Under US copyright law, only humans are deemed the authors of copyrightable works. The US Copyright Office has long opined that non-humans cannot own copyrights. See US Copyright Office, Compendium of US Copyright Office Practices § 313.2; US Copyright Office, Report to the Librarian of Congress by the Register of Copyrights 5 (1966). The Office recently adopted the same position concerning AI. See Copyright Review Board, Feb. 14, 2022 letter to R. Abbott, Esq. Re: Second Request for Reconsideration for Refusal to Register A Recent Entrance to Paradise (Correspondence ID 1-3ZPC6C3; SR # 1-7100387071), https://www.copyright.gov/rulings-filings/review-board/docs/a-recent-entrance-to-paradise.pdf. Indeed, the Copyright Office recently pulled back on its initial grant of a registration for a graphic novel whose images were generated by AI, citing potential lack of sufficient human authorship as a reason to review its decision.
Courts have taken similar positions in cases regarding non-human authors, such as the well-known “monkey-selfie case”, Naruto v Slater, 888 F.3d 418, 426 (9th Cir. 2018) (rejecting authorship to a non-human), and where “celestial beings” were alleged to have authored a book merely transcribed by humans, Urantia Found. v Maaherra, 114 F.3d 955, 958–59 (9th Cir. 1997) (human “scribes” were properly deemed the authors of the book at issue). The language of the Copyright Act is in accord, as it contains much human-centric verbiage. See – eg, 17 USC §§ 101, 201, 203, 304.
This view is currently being challenged in court by Dr Stephen Thaler, creator of an AI algorithm called Creative Machinery, who applied to copyright a generated artwork in 2018 listing the AI as author and claiming transferred ownership of the work via his ownership of the machine. The Copyright Office repeatedly denied his attempts and Dr Thaler has filed a complaint in federal court to reinstate his application and confirm his AI-as-author theory. The case may prove fertile ground for a US court to affirmatively weigh in on this issue. See Thaler v Perlmutter (1:22-cv-01564) (June 2, 2022).
End-user as author
A popular theory of authorship is that the end-user – the human at the computer terminal who enters the text “prompt” to generate the image output – should be the author even though they only wrote the words that put the AI’s wheels in motion. This approach generally tracks with copyright law’s approach to authorship of works such as photographs, where a photographer arranges the scene and then, by pressing the camera shutter, owns the photo, not the camera manufacturer or designer (let alone the camera, itself). This theory also aligns with the recent AI authorship commentary by the Copyright Office that works created solely by AI are not registerable.
But how to implement this approach in practice will be challenging given the Office’s registration practices, wherein examiners generally do not investigate or verify the information on applications. 17 USC §§ 408–410.
This approach does make sense, however, considering the AI would not independently generate artwork if not for user input, which may itself be creative if it consists of more than just a handful of words or a short phrase (for instance, a narrative paragraph or poem). See Feist Publications v Rural Telephone Service Co, 499 US 340 (1991) (threshold for copyright originality is exceedingly low). But users do not play as critical or holistic a role as, for example, photographers, given that they need only provide a short prompt, not select any lighting, angles, conditions or subjects.
In many cases, while an AI image is original enough to copyright, a user’s input may be insufficient for authorship, and given the information disclosed on the copyright application, the Copyright Office may not even know an AI created the work, what prompts were used, whether they were creative enough, and whether the user can own the resulting artwork rather than just the prompt. It seems unlikely that the Copyright Office would be able to conduct such analysis and the result could lead to ad hoc and arbitrary registrability decisions.
Developer or programmer as author
In keeping with the theme that the computer/algorithm is an inert instrument like a camera that requires human activation, one could also view the AI output as deriving largely from the developer’s software, such that the programmer (or, in many cases, the developer the programmer works for) should be deemed the author of whatever output results because arguably the output is derivative of the algorithm itself. Some developers, like OpenAI in its terms of use for Craiyon, have deemed themselves owners of the engine’s output and grant a license to end-users to make certain uses of the images, whereas others, such as Midjourney, state in their terms of use that the end-user is the owner of the copyrights in generated images, but grants a license to Midjourney.
This approach would imbue companies with vast control over innumerable works, including ones that have not yet been created, while ignoring the village it takes to facilitate AI output and the large dataset selected for AI training. Nor have developers and programmers in other industries historically received ownership in this way.
Moreover, courts have commonly dismissed programmers’ attempts to assert ownership of the programs’ output. For instance, in Rearden LLC v Walt Disney Co, 293 F. Supp. 3d 963, 967–71 (N.D. Cal. 2018), a software company behind facial performance capture software used in films sued for copyright infringement, alleging that studios had unauthorisedly used its technology, and that it owned the software’s output. The court dismissed these claims, saying that software owners’ copyrights should only extend to program output where the program “does the lion’s share of the work” and the user plays a marginal role. Id. at 970-71. This holding, however, revives the question of whether an AI user’s input is marginal or not, and what constitutes “the lion’s share of the work”. Perhaps as to AI, courts could distinguish this early case law, holding that AI systems arguably do the lion’s share of the work, with minimal input from users. Coming full circle, an analysis of “lion’s share” will be fact-intensive and may lead to arbitrary results, whether by courts or the Copyright Office.
Hybrid contractual approach
Under US law, one can transfer ownership of a work by a signed written agreement. See 17 USC §§ 201(d); 204(a). One potential solution to the authorship/ownership quandary – albeit voluntary – would be to confer joint authorship between end-users of AI platforms and developers. This would account for the fact that both parties are, in varying degrees, necessary to an AI-generated work.
While facially appealing, this approach may also be challenging to implement. Along with a signed writing, co-authors must intend that their contributions merge into a single work. See 17 USC § 101. Simply adding this language to boilerplate terms and conditions may not suffice, see Thomson v Larson, 147 F.3d 195, 201 (2d Cir. 1998), and asking users to review and sign a formal contract may be logistically burdensome and deterrent. Moreover, developers would lack incentive to enter such arrangements when they could claim full ownership from the outset, as many have through their terms of use.
The public domain – no copyright protection
Another proposal is that of no ownership at all. AI-generated works could potentially be considered owned by no one, given the lack of significant human input (particularly where just a few keywords are used). Such works – which are often shared through social media, AI websites and Discord channels – would then be free for anyone to use. However, this view conflicts with some platforms’ terms, which claim the developer or user is the owner.
The question of authorship and copyrightability of AI-generated works is in flux, likely with no clear answer until courts issue decisions interpreting the current law as applied to generative AI or legislation amending the Copyright Act is passed, accounting for this new technology. Either avenue is a slow process, and meanwhile AI will continue to advance and outpace the law.
Infringement and fair use
Training data
Beyond authorship, generative AI raises significant concerns regarding whether its unauthorised scraping of training data constitutes infringement or is protected by fair use, and whether its output is infringing of existing works, particularly where users can ask AI to emulate specific artists’ styles. Training data is often scraped from the web, as in the case of text-to-image AI such as Stable Diffusion, whose dataset comprised images scraped from hundreds of online domains, including WordPress, Shutterstock, and Getty Images.
Copyright infringement issues arise where the training content contains copyrighted works used without authorisation, as copying this content violates the exclusive reproduction right under the Copyright Act. See 17 USC § 106. Whether the practice of scraping data should be excused as fair use is as yet unanswered. See id. § 107. Fair use permits unlicensed uses of copyrighted works in certain contexts to promote further expression and innovation – including criticism, commentary, news reporting, teaching, scholarship, and research – based on an analysis of four non-exclusive statutory factors.
The first addresses the purpose and the character of the use, including its commercial nature, and whether the new work transforms the original, imbuing it with a further purpose or different character. This factor, for years, has been considered paramount, particularly the question of “transformative use”, which the Supreme Court adopted as a measure of whether a secondary use of a copyrighted work merely supersedes the objects of the original creation. Campbell v Acuff-Rose Music, Inc, 510 US 569, 579 (1994).
Whether using scraped data to train an AI may be considered transformative (as opposed to, for instance, use of online images for aesthetic or conceptual purposes) will be a key question, and the Supreme Court’s forthcoming decision in Warhol v Goldsmith may illuminate how this factor should be analysed. Proponents may argue that training sets contain transformative purposes, while opponents would argue that AI’s task is not transformative because it is using existing works to create new visual works for similar and often competing purposes, such as for art, conceptual design or entertainment.
The second fair use factor addresses the nature of the copyrighted work, including whether it is a creative versus factual work. This factor typically is not seen as determinative, but when a work is more factual, the scale tips in favour of fair use, compared to strictly creative works.
The third factor addresses the amount and substantiality of the copyrighted work used, which in the case of AI training, would typically be an entire work. With visual art, courts have allowed the use of entire underlying works where it is transformative and does not supplant the original. See – eg, Bill Graham Archives v Dorling Kindersley Ltd, 448 F.3d 605, 613 (2d Cir. 2006) (“[C]opying of an entire work . . . does not necessarily weigh against fair use because copying the entirety of a work is sometimes necessary to make a fair use of the image.”).
The fourth factor examines potential market harm the secondary work could cause for the rightsholder’s ability to exploit an original work. This factor used to be seen as paramount, given the economic ramifications for plaintiffs, and asks whether existing or likely licensing markets for the underlying work will be impacted. How this factor tips will largely depend on the nature of the underlying work and artist, and whether there is – or could be – a market for licensing content for use to train AI. Current industry practices shed some light, as image licensing company Shutterstock entered into a licensing arrangement with DALL-E 2, underscoring the value in the visual content and metadata associated with image files.
Considering the emerging nature of AI, US courts have not yet had an opportunity to address the legality of AI datasets. At the outset, it may be difficult for an author to discover that their work was used for training purposes, considering the sheer volume of data involved, as well as secrecy around the data used. However, case law is on the horizon. Concerned visual artists have begun filing suit, including a recent class action in California, Anderson v Stability AI Ltd (3:23-cv-00201), raising infringement and other claims against Stability AI, Midjourney and DeviantArt for unauthorisedly using the artists’ original works in training sets that allow end-users to replicate artworks in their style. Additionally, visual content licensing platform, Getty Images, recently began proceedings against Stability AI in London, asserting infringement of millions of scraped images owned or represented by Getty Images. A separate class action brought by a programmer accuses Microsoft, GitHub and OpenAI of unauthorisedly reproducing open-source code through AI coding assistant Copilot. This case addresses coding and not text-to-image generators, but the general notion of using existing copyrightable works as training data will likely be front and centre.
Sensing the coming storm, many generative AI companies have already begun implementing policies to safeguard against infringement claims. Many incorporate terms of service that require users to avoid using prompts that draw from copyrighted works to generate works, which, admittedly, is hard to enforce. Dataset licensing practices are also emerging across industries, though detractors argue that licensing at this scale is impractical and could hinder training.
The image licensing industry is also adopting varying approaches to generative AI and image licensing. Adobe Images has guidelines for submitting images created using generative AI, requiring the images be tagged as AI and restricting any works depicting real places, identifiable property (such as famous characters or logos) or notable people (whether photorealistic or caricatures). Shutterstock has partnered with OpenAI (DALL-E 2) to allow customers to create and use AI images, under terms and conditions. Some companies are creating their own datasets to create synthetic images such as vAIsual, a technology company building AI synthetic images created with a licensable biometrically-released real-life dataset.
Other companies have created or adopted metadata image tags to establish that particular images should not be scraped as data. Finally, insurance and indemnity agreements with image licensing companies may help users reduce the risk of using some AI-generated content for commercial, rather than personal use.
Output
Claims of copyright infringement may also arise when the output from AI generators appears substantially similar to existing works, by virtue of those works being included in the training data or where users ask the generator to create a work in a specific artist’s style. When a resulting work – often freely viewable online – appears similar to an existing work, a threshold question for a putative plaintiff will be who to sue. The user, developer and programmers are all potential targets.
One of the most reported examples of an individual artist’s works being copied by users by incorporating his name into prompts is Greg Rutowski, an illustrator whose work has appeared in games like “Dungeons & Dragons” and who now faces a parade of AI-generated works that are available online in his style and falsely attributed to him, making it difficult to find his real artwork. Similarly, a Disney illustrator recently discovered that her style had been cloned by an AI, which was fed 32 pieces of her downloaded artwork, without her permission. While an artist’s “style” is not protected by copyright, many visual artists raise concern that text-to-visual generative AI are mimicking their style and expression, raising ethical as well as copyright questions.
Some AI platforms and online art communities, including ArtStation, DeviantArt and Stable Diffusion, have adopted measures to allow artists to opt-out and remove their works from future training datasets. While a noteworthy concession, many artists support an affirmative opt-in, with the default being that their works cannot be used; technologists disagree, stressing that seeking affirmative permission from every artist with work on the internet would be impossible. There are sites that are even dedicated to helping artists, such as photographers, learn if their images are included in training sets, such as haveibeentrained.com.
Conclusion
Generative AI is still in its nascent phase, with artists, licensors, technology companies and users navigating both the potential for creativity as well as the harm that arises when creating new works. Visual artists legitimately fear that new works created in their style will compete with their copyrighted works. Or worse, that works will be attributable to them that they did not create. Who will be legally responsible for any misappropriation is uncertain.
At the same time, creators of AI generative art may not be able to own copyright in the output. Companies are grappling with the capabilities of generative AI and attempting to develop policies that can take advantage of the new possibilities, while considering the ethical and legal issues involved with the creation of synthetic content. The current copyright system may or may not be adequate to answer these questions, leaving the courts and Congress to determine how to balance the rights of all stakeholders.
41 Madison Avenue
Floor 38
New York
NY 10010
USA
+1 212 974 7474
+1 212 974 8474
nwolff@cdas.com www.cdas.com