AI and Copyright: What Businesses Need to Know in 2025
Artificial intelligence is transforming how companies create and deliver value. Businesses now use AI to draft contracts, write code, design marketing materials, and even develop new products. Technology is moving fast, but the law has not kept pace. One question has become central: who, if anyone, owns the rights in AI-generated output? U.S. law is moving toward an answer, though not yet settled. Courts and the Copyright Office agree on one core principle — copyright protection still depends on human authorship.
1. The Statutory Framework
The Copyright Act of 1976 protects “original works of authorship” fixed in a tangible medium. The Act doesn’t define “authorship,” but courts have long read it to mean a human creative act.
Constitutional foundation
The Constitution’s Intellectual Property Clause empowers Congress to promote “the Progress of Science and useful Arts” by securing rights “to Authors.”
As the Supreme Court explained in Feist Publications v. Rural Telephone Service (1991), this language contemplates human creators and seeks to encourage human intellectual advancement.
The principal dates to Burrow-Giles Lithographic Co. v. Sarony (1884). There, the Court upheld copyright in a photograph of Oscar Wilde because the photographer’s choices, lighting, pose, composition, showed human authorship. Even where technology assists, copyright still requires identifiable human control over expressive elements.
2. Key Cases in the Second Circuit
For companies in New York, Connecticut, and Vermont, Second Circuit precedent governs. Two decisions are particularly relevant to AI.
Cartoon Network LP v. CSC Holdings (“Cablevision”), 536 F.3d 121 (2d Cir. 2008).
The court held that direct infringement requires volitional human conduct. Because users, not Cablevision, initiated DVR recordings, the company wasn’t directly liable. The Second Circuit reaffirmed this in Capitol Records v. ReDigi (2018), clarifying that even automated processes require human volition to initiate copying.
The takeaway: if no human makes the creative or volitional act of authorship, an AI system’s output may fall outside copyright protection.
Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015).
While upholding Google’s book digitization as fair use, the court stressed that copyright exists to incentivize human creativity while permitting innovation that benefits the public.
The same tension between innovation and authorship now defines the AI debate. Together, these decisions suggest that the Second Circuit is unlikely to recognize copyright in fully autonomous AI output.
3. Copyright Office Guidance
The Copyright Office’s 2023 guidance set three clear principles:
Works generated entirely by AI are not registrable.
Hybrid works may qualify, but only for the human-authored portions.
Applicants must disclose AI involvement when seeking registration.
This policy was upheld in Thaler v. Perlmutter (D.D.C. 2023), where the court refused to register AI-generated artwork, calling human authorship “a bedrock requirement of copyright.”
The Zarya of the Dawn example
In 2023, the Office addressed a comic book created with Midjourney.
It registered the work but excluded the AI-generated images, protecting only the human-authored text, selection, and arrangement.
The message for businesses: document your human creative contributions.
Pending litigation
Cases such as Andersen v. Stability AI, Authors Guild v. OpenAI, and Getty Images v. Stability AI will test fair use and training issues, but they are unlikely to settle the authorship question.
4. What This Means for Business
Pure AI works lack copyright protection.
If an output is generated autonomously, competitors may reuse it freely.
This risk is most acute in marketing, software, and design industries that rely heavily on generative models.
Hybrid works may be partially protected.
When humans contribute creative direction, editing, or curation, those elements can qualify as original authorship.
Following the Zarya model, record human contributions carefully.
Trade secrets become vital.
Where copyright fails, trade secret protection may cover training methods, datasets, and proprietary processes. Confidentiality may preserve more value than publication.
Contracts matter more than ever.
Until Congress or the courts provide new rules, contracts are the most effective protection:
Define ownership of AI outputs and underlying data in vendor agreements.
Use licenses and NDAs to limit reuse or disclosure.
Capture evidence of human contribution in internal policies.
Address AI tool limitations and ownership gaps in service agreements.
Practical steps
Audit current AI use and identify protection gaps.
Revise vendor contracts to clarify ownership of generated content.
Train teams to document human creative contributions.
Consider trade secret protection for proprietary models.
Monitor regulatory developments as oversight expands.
5. Conclusion
AI has changed how companies create and compete, but copyright law still draws the line at human creativity.
Courts continue to require human authorship, grounded in constitutional text and policy.
The message for business is simple: do not rely on copyright alone to protect AI-generated works.
Use contracts, trade secrets, and internal governance to secure ownership and preserve value.
Organizations that document human involvement and negotiate clear contractual rights will be best positioned as this landscape evolves.
(There is no “20 percent human contribution” rule. What matters is the nature and quality of human creative control, not a percentage threshold. When registering works, applicants should describe their human contributions clearly.)
Gary Lipson, Esq.
New York attorney focusing on intellectual property, licensing, and strategic transactions.
This article is for informational purposes only and does not constitute legal advice.