Publishers sued OpenAI for training ChatGPT on copyrighted articles.
Artists sued Stability AI for training image generators on copyrighted artwork.
Authors sued multiple AI companies for using books without permission.
These lawsuits target AI vendors. But they won’t stop there.
The next wave targets companies using AI trained on copyrighted content.
Most companies have no idea the AI they’re using was trained on material they don’t have the rights to.
The Copyright Lawsuits Against AI Vendors
Since 2023, major copyright holders have sued AI companies, including The New York Times v. OpenAI, Getty Images v. Stability AI, Authors Guild v. OpenAI, and record labels v. AI music generators.
The claims are consistent: AI companies scraped copyrighted content without permission, used it to train AI models, created systems that compete with the original content, and profited from copyrighted material without compensating owners.
AI companies argue for fair use. Copyright holders argue it’s a massive infringement.
Courts are starting to rule. Early decisions are mixed. Some allow cases to proceed. Some dismiss portions of the claims.
But one thing is becoming clear: using copyrighted content to train AI isn’t automatically legal.
Why This Affects Companies Using AI
Companies think that “I’m just a customer, not the AI vendor” protects them.
It doesn’t.
Secondary liability: Copyright law includes contributory infringement and vicarious liability. Companies using AI tools that infringe copyright can be liable for the infringement.
Direct infringement: When companies use AI to generate content that’s substantially similar to copyrighted works, they directly infringe even if the AI vendor did the training.
No fair use safe harbor: Fair use is a defense, not a right. The fact that the AI vendor claims fair use doesn’t protect customers if courts reject that defense.
Indemnification won’t help: AI vendor contracts include indemnification clauses, but they’re limited, exclude certain claims, and may be worthless if the vendor can’t pay or goes bankrupt.
Companies using AI trained on copyrighted content inherit the copyright risk.
Real Exposure Scenarios
Marketing content: The company uses AI to generate blog posts, social media content, and marketing materials. The AI was trained on copyrighted articles and generates content substantially similar to the training data. Copyright holders sue both the AI vendor and the company for infringement.
Code generation: The company uses AI to write software code. The AI was trained on copyrighted code repositories. It generates code that’s substantially similar to copyrighted code. The code owner sues for copyright infringement.
Image generation: The company uses AI to create product images and designs. The AI was trained on copyrighted photographs and artwork. It generates images that copy elements of the training data. Artists and photographers sue.
In each case, the company argues they’re just using a tool. Courts may still find them liable for the infringing output.
The Due Diligence Problem
When companies deploy AI, they should ask the vendor what data the AI was trained on, whether the vendor has rights to use that training data, what copyright clearance process was followed, and what happens if training data is found to be infringing.
Most vendors can’t or won’t answer these questions clearly.
They say training data is “publicly available information” (which doesn’t mean they have rights to use it), the training process is “proprietary” (which avoids answering the copyright question), or they claim “fair use” (which isn’t established law yet).
Companies deploy AI without understanding the copyright risk in the training data.
What Courts Are Saying
Early court decisions are mixed, but several patterns are emerging.
Fair use isn’t automatic: Courts aren’t automatically accepting AI companies’ fair use defense. They’re analyzing it case by case.
Commercial use matters: AI companies’ commercial use of copyrighted material weighs against fair use.
Market substitution matters: If AI-generated content substitutes for the copyrighted original, that weighs against fair use.
Wholesale copying matters: Training on entire copyrighted works without selection or limitation weighs against fair use.
The outcome isn’t certain, but the risk is real.
What Companies Should Do
If companies use AI to generate content, code, images, or other potentially copyrighted output, they need to understand the risk.
Ask vendors about training data: Require transparency about what data was used to train the AI.
Assess copyright risk: Evaluate whether AI-generated output could infringe copyrighted works.
Review indemnification: Understand what AI vendor indemnification actually covers and its limitations.
Implement review processes: Have humans review AI-generated output for potential copyright issues before use.
Consider alternatives: For high-risk uses, consider AI trained on licensed content or proprietary data rather than scraped copyrighted material.
Monitor litigation: Track major AI copyright cases because outcomes will affect customer liability.
Most companies do none of this because they assume the AI vendor handled copyright clearance.
Schedule a 30-minute AI Copyright Risk Assessment where we’ll:
- Review what AI systems may have copyright exposure
- Show what due diligence questions to ask vendors
- Assess indemnification protection gaps
- Create review processes for AI-generated content
- Help evaluate alternatives for high-risk AI uses
Because using AI trained on copyrighted content feels safe until the copyright lawsuit names your company, too.
Brian Alexander
Founding Partner, VisionAI+ Consulting Group
Former General Counsel, AI Engineering Firm
Certified ISO 42001 Lead Auditor