US Copyright Law & AI-Generated Content
A comprehensive analysis of what US copyright law actually says about AI-assisted creative works
Executive Summary
β Copyrightable
Tool use with human creative control
- Human makes creative decisions
- Iterative refinement and selection
- Arrangement and composition
- Post-generation modifications
- Integration into larger work
β Not Copyrightable
Autonomous AI generation without human input
- AI listed as sole author
- Zero-shot generation
- No human selection process
- Pure delegation to AI
- Unable to document human contribution
Constitutional & Statutory Basis
Copyright Clause (Article I, Section 8, Clause 8)
"To promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries"
Key term: "Authors" - courts have historically interpreted this as requiring human creators
Copyright Act of 1976 (17 U.S.C.)
- Β§102(a): "Copyright protection subsists...in original works of authorship"
- Β§201(a): "Copyright in a work...vests initially in the author or authors of the work"
Major Court Cases
Thaler v. Perlmutter (D.C. Circuit, March 18, 2025)
Facts:
- Dr. Stephen Thaler created "Creativity Machine" AI system
- AI autonomously generated artwork "A Recent Entrance to Paradise"
- Thaler listed the AI as sole author, himself as owner via work-for-hire
- Copyright Office denied registration
Holding:
The DC Circuit AFFIRMED denial, holding that:
"The Copyright Act of 1976 requires all eligible works to be authored in the first instance by a human being."
β What the case ACTUALLY says:
- AI cannot be listed as an author
- The work was created ENTIRELY by AI with no human creative input
- Thaler explicitly disclaimed human authorship to make a test case
β What the case DOES NOT say:
- That humans using AI tools lose copyright
- That AI assistance eliminates copyrightability
- That prompting or directing AI counts as non-human authorship
- That looking at copyrighted code with AI makes output derivative
Critical Waiver: Court noted Thaler WAIVED his alternative argument that he should be considered author by virtue of creating/using the machine.
Zarya of the Dawn (Copyright Office, February 21, 2023)
Facts:
- Kristina Kashtanova created graphic novel using Midjourney AI
- Initially registered without disclosing AI involvement
- Copyright Office discovered AI use and reassessed
Outcome:
PARTIAL REGISTRATION APPROVED
What IS protected:
- Written text (human-authored)
- Selection, coordination, and arrangement of visual elements
- Overall compilation and structure
What is NOT protected:
- Individual Midjourney-generated images
- AI-generated visual content itself
"The fact that Midjourney's specific output cannot be predicted by users makes Midjourney different for copyright purposes than other tools used by artists."
Key distinction: Insufficient creative control over specific AI outputs vs. creative arrangement of those outputs
Theatre D'opΓ©ra Spatial (Review Board, September 5, 2023)
Facts:
- Jason Allen created award-winning artwork using Midjourney
- Used 624+ prompt revisions
- Applied Photoshop edits and Gigapixel AI upscaling
- Initially didn't disclose AI involvement
Outcome:
REGISTRATION DENIED
Why it was denied:
- Allen REFUSED to disclaim AI-generated portions
- Work contained "more than de minimis" AI content
- Allen claimed copyright in AI-generated elements themselves
What Allen should have done:
- Disclaim AI-generated base imagery
- Claim only human modifications and arrangements
- Provide documentation of creative process
Approved Registrations - Proof It Works
Invoke "American Cheese" (January 30, 2025)
BREAKTHROUGH CASE - First image composed entirely of AI-generated material to receive copyright
Facts:
- Kent Keirsey (Invoke CEO) created image titled "A Single Piece of American Cheese"
- Used Invoke's inpainting features
- Made ~35 iterative edits
- Documented entire creative process
Why it succeeded:
- β Demonstrated human creative control through iterative editing
- β Selection and arrangement of AI-generated elements
- β Comprehensive provenance documentation
- β Showed decision-making at each step
What was protected:
"Selection, coordination, and arrangement" of AI-generated elements - NOT the individual AI outputs
Key innovation:
Provenance Records tool tracking every creative decision
Raksha World Ltd. (CopySight AI, 2024)
Status: Approved as filed
Significance: Demonstrates that properly documented AI-assisted works can receive full copyright protection
Key factor: Framework establishing human authorship through documented creative process
Copyright Office Guidance Evolution
March 16, 2023 - Policy Statement on AI
Key Requirements:
- Applicants MUST disclose AI-generated material exceeding de minimis amounts
- Must provide brief explanation of human author's creative contribution
- AI-generated traditional elements of authorship lack copyright protection
Copyrightable Scenarios:
- β Human selects/arranges AI material in sufficiently creative ways
- β Human modifies AI output to degree meeting copyright standards
- β AI used as assistive tool within larger human creative process
January 29, 2025 - Part 2 Copyrightability Report
What Does NOT Bar Copyrightability:
- β Using AI as a tool in creative process
- β Including AI-generated material in larger human-created work
- β Human modifications or arrangements of AI outputs
What IS Required:
- Human author must determine sufficient expressive elements
- Human-authored work must be perceptible in output
- Creative arrangements/modifications by humans
However: Prompts + selection + iteration + arrangement + modification CAN establish human authorship
What Counts as "Sufficient Human Creative Control"
Based on Copyright Office guidance and case outcomes:
SUFFICIENT:
- β Iterative editing and refinement
- β Selection among multiple AI outputs
- β Arrangement and composition decisions
- β Post-generation modifications
- β Integration into larger human-created work
- β Creative direction affecting multiple expressive elements
- β Using AI to implement specific creative vision
INSUFFICIENT:
- β Single zero-shot prompt with no follow-up
- β Accepting first AI output with no selection
- β Minimal or no post-generation changes
- β Pure delegation of creative decisions to AI
- β Unable to document human creative contributions
The Critical Word: "Autonomous"
From Thaler opinion and Copyright Office guidance:
"Autonomous" = AI generating work without human creative control
The Spectrum of Human Control:
[No Copyright] β------------------------------------------------β [Full Copyright]
Fully autonomous Some human Significant AI as assistive
AI generation selection human control tool in human
creative process
Documentation Requirements
Based on successful registrations (especially Invoke):
What to Document:
- Creative process: Step-by-step decision-making
- Iterations: Multiple attempts showing human direction
- Selection criteria: Why specific outputs were chosen
- Modifications: Post-generation human edits
- Arrangement: How elements were composed
- Tools used: AI systems, versions, features
- Timestamps: Chronological creation record
- Git commits: For code, detailed commit history
Provenance Records Model:
- Track every AI model invocation
- Log all parameters and settings
- Record all human decisions
- Save intermediate outputs
- Document selection rationale
Common Misinterpretations - CORRECTED
β WRONG: "Thaler case means AI tools can't be used"
β CORRECT: Thaler only addressed AI as sole author; explicitly didn't address AI assistance
β WRONG: "Any AI involvement eliminates copyright"
β CORRECT: AI as tool with human creative control retains copyright
β WRONG: "Prompting = no human authorship"
β CORRECT: Prompting ALONE insufficient, but prompting + selection + iteration + arrangement CAN establish authorship
β WRONG: "Copyright Office won't register AI-assisted works"
β CORRECT: Multiple successful registrations prove otherwise (Invoke, Raksha World, etc.)
β WRONG: "Training data in LLM taints all output"
β CORRECT: Training and generation are separate; output originality assessed independently
β WRONG: "Clean room implementation can't use AI"
β CORRECT: Clean room is defense strategy, not copyright requirement; AI can assist research
Key Takeaways for Developers
Using GitHub Copilot, Claude, ChatGPT, etc.:
You retain copyright IF:
- You make creative decisions about what code to use
- You modify and adapt AI suggestions
- You integrate AI outputs into your larger codebase
- You can document your creative contributions
- Your git history shows iterative human development
Document:
- Commit messages explaining your decisions
- Code reviews showing human judgment
- Modifications to AI suggestions
- Architecture and design decisions
References & Primary Sources
Primary Legal Documents:
- Thaler v. Perlmutter, No. 23-5233 (D.C. Cir. Mar. 18, 2025)
- U.S. Copyright Office, Copyright Registration Guidance: Works Containing Material Generated by AI (Mar. 16, 2023)
- U.S. Copyright Office, Copyright and AI Part 2: Copyrightability (Jan. 29, 2025)
- Zarya of the Dawn Registration Letter (Feb. 21, 2023)
- Theatre D'opΓ©ra Spatial Review Board Decision (Sept. 5, 2023)
Key URLs:
Last Updated: March 2026