If your clients use artificial intelligence (“AI”) tools for content creation, something that is incredibly likely given AI’s widespread adoption of late, two federal court decisions from June 2025—Kadrey v. Meta Platforms, Inc.[1] and Bartz v. Anthropic PBC[2]—just changed their liability exposure. Both rulings found that AI training constitutes fair use, but each emphasized that businesses using AI-generated content remain liable for any copyright infringement in the outputs that they create and publish.
These cases focused on AI training specifically, but, in doing so, they also addressed output liability—highlighting risks that many businesses haven’t fully considered or prepared for. For example, a marketing firm using AI to draft client proposals faces the same copyright liability that it would for human-authored content that infringes, but it may not have policies addressing this exposure. Similarly, a law firm generating brief templates with AI faces potential copyright issues that don’t exist when using AI for case research. Understanding how these decisions analyze AI technology helps business lawyers better advise clients on managing copyright risks in all aspects of AI-generated content creation.
What the Courts Actually Decided
Both courts distinguished between AI training (how systems learn from data) and AI generation (when users prompt systems to create content). While training received fair use protection, the judges emphasized that businesses using AI tools remain liable for copyright infringement in the outputs that they create and use.
Judge Chhabria in Kadrey introduced a “market competition” theory that could dramatically expand copyright liability. Traditional copyright law requires that infringing works serve as substitutes for originals—someone reads the copy instead of buying the original. The Kadrey court suggested that AI-generated content might infringe simply by competing with copyrighted works, even without direct substitution. Under this theory, a consulting firm using AI to generate market research reports could face liability if those reports compete with copyrighted research, regardless of whether clients would have purchased the original reports.
Judge Alsup in Bartz took a different approach, fragmenting the analysis into separate questions about data acquisition and training. His decision suggests that how AI companies obtained their training data could affect fair use protection for end users. Companies using AI systems trained on unauthorized content might face greater liability than those using systems trained only on licensed materials—though most businesses have no visibility into their AI providers’ training data sources.
Immediate Business Risk Assessment
These decisions create different exposure levels depending on how clients use AI tools.
Content creation presents the highest risk under the new Kadrey framework. An advertising agency using AI to create campaign materials that resemble existing advertisements faces potential liability even if the AI outputs serve different purposes than the originals. A marketing firm generating social media content with AI tools must consider whether those outputs compete with copyrighted posts, graphics, or campaign materials.
Professional services face more complex analysis. A law firm using AI to draft contracts or briefs creates potential exposure if outputs closely resemble copyrighted legal materials, particularly specialized forms or distinctive arguments from legal publications. However, a firm using AI for case research or document review operates in safer transformative use territory because these applications serve different purposes than the original materials.
Healthcare organizations using AI for patient communications or educational materials must monitor whether outputs resemble copyrighted medical publications or patient education resources. Financial services firms generating investment analysis or client reports with AI face liability if those outputs compete with copyrighted financial research or proprietary investment strategies.
Operational applications like customer service chatbots or internal documentation generally present lower risks, but companies should still establish policies preventing deliberate reproduction of copyrighted materials.
Five Critical Questions Before Using AI for Content Creation
Business lawyers should walk clients through this assessment before implementing AI content generation:
- Does the AI output compete in the same market as copyrighted works? If yes, document why the use serves different purposes and implement review procedures.
- Can you identify the training sources for your AI system? If using commercial AI services with opaque training data, which will often be the case, strongly consider additional safeguards against reproducing copyrighted content.
- Do you have policies preventing deliberate copying? Establish clear guidelines prohibiting employees from prompting AI systems to reproduce known copyrighted materials.
- Can you demonstrate transformative purpose? Document how AI usage serves legitimate business functions distinct from consuming original copyrighted works.
- Do you have review procedures for high-risk outputs? Implement screening for AI-generated content intended for publication, marketing, or external distribution.
Practical Compliance Framework
Companies should implement documentation showing good-faith efforts to prevent copyright infringement. This includes maintaining records of AI implementation purposes, establishing clear usage policies that prohibit deliberate reproduction of copyrighted content, and providing employee training on both AI capabilities and copyright limitations.
For higher-risk applications like content creation, establish review procedures before publication or distribution. A simple workflow requiring human review of AI-generated marketing materials, social media posts, or client communications can demonstrate reasonable efforts to prevent infringement while preserving operational benefits.
Employee training should emphasize that fair use protection isn’t automatic. Staff need to understand that prompting AI systems to create content “in the style of” specific copyrighted works or asking for materials that closely mimic known publications creates liability exposure. Clear examples help: asking ChatGPT to “write a blog post about cybersecurity” generally presents low risk, while asking it to “write a blog post like the recent Harvard Business Review article on cybersecurity” creates potential problems.
Companies should also address inherited liability from AI systems trained on questionable datasets. While legal standards remain unclear, businesses can demonstrate good faith by avoiding AI services known to have used unauthorized training materials and transitioning to providers with transparent data sourcing when feasible.
Documentation should include regular risk assessments evaluating AI applications across different business functions. Higher-risk uses like creative content generation require additional safeguards, while operational applications like data analysis face lower exposure.
Litigation Strategy Implications
These recent court decisions shift copyright litigation strategy significantly. Discovery will focus on internal policies, employee training, and evidence of intentional copying rather than just comparing AI outputs to copyrighted works. Companies that can show robust compliance frameworks and good-faith efforts to prevent infringement strengthen their defense position.
The Kadrey market competition theory creates new motion practice challenges. Even businesses demonstrating transformative purpose might face factual disputes about market competition that survive early dismissal motions. This makes settlement discussions more attractive, particularly when focused on prospective compliance measures rather than retrospective damages.
Defense strategies should emphasize documented transformative purposes and compliance with recognized governance frameworks. Companies maintaining clear records of legitimate business purposes for AI usage, comprehensive employee training, and procedures for addressing potential copyright issues will have stronger positions in any litigation.
Managing Uncertainty During Appeals
While these district court decisions will likely face appellate review, the two- to three-year timeline for resolution means that businesses must address current exposure. The cost of implementing reasonable AI governance policies remains modest compared to potential copyright litigation expenses, making cautious compliance a sound business decision regardless of how these legal theories ultimately develop.
Companies should monitor legal developments while implementing protective measures now. Basic documentation, employee training, and usage policies provide protection against various liability theories, not just the specific approaches in Kadrey and Bartz. These measures also demonstrate good faith in any future litigation, providing valuable leverage regardless of how appellate courts rule.
The key insight from these decisions is that businesses cannot simply assume that AI tool usage is protected. While AI training generally receives fair use protection, companies using AI-generated content must implement appropriate safeguards and compliance measures. Those establishing robust governance frameworks now—treating these decisions as important guidance while they work through the appellate process—will be best positioned regardless of how this legal landscape ultimately develops.
