Today´s discussion
AI and Copyright Law: Navigating the New Frontier of Digital Rights
PROTECTING CREATIVITY'S NEW FRONTIER
Can a machine-generated artwork be protected by copyright ? As artificial intelligence reshapes creative industries, this question becomes increasingly critical for creators, businesses and legal frameworks worldwide.
Generative AI is creating new questions about who owns creative work. As AI tools can now write, draw and compose, we need to figure out who has legal rights over AI-created content.
In the past, copyright laws only protected things people made with their own hands or minds : like books, buildings or songs. Now, as we live more of our lives online, copyright also protects digital creations. This includes computer programs, e-books, online articles and websites. The law had to grow to include these new types of creative work to make sure people's ideas are protected, no matter if they exist on paper or on a screen.
Which Law and policy Considerations for AI Copyrighting ?
When you create something original - whether it's art, music, writing, or other creative work - copyright protection begins automatically in most countries that follow the Berne Convention rules. You don't need to register or file paperwork to have basic copyright protection. However, some countries like the United States offer optional registration that can provide additional legal benefits if you need to defend your rights in court.
Copyright owners have 2 main rights:
- Economic rights - letting them earn money by allowing others to use their work through licenses
- Moral rights - protecting non-financial interests, such as being recognized as the creator and preventing changes that could damage their reputation
Copyright law protects creative works by making it illegal for others to use them without permission.
This includes:
- Copying the work
- Changing or adapting it
- Sharing or distributing it
- Performing it in public
- Creating new works based on it
Using someone's work without their permission is called copyright infringement.
If you want to use someone's copyrighted work, you generally need their permission first. However, some countries have "fair use" or "fair dealing" laws that let you use copyrighted material without permission in specific situations. These exceptions limit the copyright owner's control over their work, making it possible to use content legally for certain purposes without asking first.
→ In the United States and Singapore, courts use specific factors to decide if using copyrighted work counts as "fair use":
- Why and how the work is being used
- What type of copyrighted work it is
- How much of the work is being used
- How the use might affect the original work's market value
Singapore previously had a 5th factor: whether the work could be purchased at a normal price within a reasonable time. Under Singapore's new 2021 law, courts only consider this when relevant.
→ The UK allows "fair dealing" as an exception to copyright law, though it's not strictly defined.
Courts look at each case individually, considering:
- How the use affects the original work's market
- If the amount used is reasonable
- The specific facts of each situation
Unlike the US system, UK courts decide what counts as fair dealing case by case rather than following fixed rules.
When courts rule on copyright violations, they can require several penalties. They may order the infringer to stop using the work immediately and pay for damages caused. Courts can also set specific financial penalties defined by law or make the infringer hand over any profits they made. Additionally, courts often order illegal copies to be destroyed and require the infringer to cover all legal fees and costs related to the case.
Examples of Generative AI Copyright Litigation in the U.S.
In the United States, several major copyright lawsuits have emerged around generative AI.
Getty Images sued Stability AI over allegedly training Stable Diffusion on copyrighted images without permission. Authors like Sarah Silverman filed class action lawsuits against Meta and OpenAI, claiming their books were used to train language models without authorization.
The New York Times sued OpenAI and Microsoft over using its articles to train ChatGPT. The Times claims these companies copied their journalism without permission to build AI tools that now compete with them as information sources. OpenAI counters that this use falls under "fair use" - a legal principle that has historically protected new technologies that transform existing work in innovative ways. They argue copyright law wasn't meant to block technological advancement that builds upon existing works.
These cases center on whether using copyrighted material to train AI models constitutes fair use. The outcomes could significantly impact how AI companies gather training data and whether they need licenses from copyright holders.
Copyright holders challenging AI companies have 2 main arguments. First, they claim AI companies illegally copied protected works to train their models. Second, they argue that AI-generated content trained on copyrighted material counts as unauthorized derivative works.
Courts examining fair use in AI cases must consider several key factors: what exactly was copied, whether using copyrighted material for AI training counts as "transformative use," how much material was copied and how AI-generated content impacts the original work's market value.
Another example is the recent California court case, Tremblay v. OpenAI, addressed these issues. Authors sued OpenAI for using their works without permission or payment to train AI models. While the court dismissed some claims about copyright violations and unfair enrichment, the core question remains: does training AI on copyrighted works require permission from creators ?
On the other side, Microsoft has taken steps to protect its Copilot customers from legal risks related to AI-generated content. The company promises to take responsibility if their commercial customers face legal problems from using Copilot. However, customers must follow certain rules: they must use content filters and safety features built into the product and they can't use inputs they don't have rights to when generating content.
Is Citing Source a Good Practice ?
Citing sources properly when using others' content is both legal and considered best practice in academic and journalistic writing. Under "fair use" principles, you can reference and quote from existing works if you:
- Properly cite the original source
- Use reasonable portions of the content
- Add your own analysis or perspective
- Don't negatively impact the original work's market
This differs from AI training data usage where entire works are processed without explicit attribution. For professional content, adding a "Sources" or "References" section demonstrates transparency and respect for intellectual property while protecting yourself legally.
Example citation format: "As discussed in [Author's] report on [Topic] (2023), [specific point]..." Followed by full citation in a references section.
How to Implement AI Governance for AI Copyrighting ?
As copyright lawsuits around AI continue, organizations are developing strategies to protect themselves and respect intellectual property rights with copyright safety solutions.
These include:
→ Getting clear permission from content owners before using their work for AI training, implementing content filters to prevent copyright violations, documenting AI system training data sources, developing clear policies for handling copyrighted material and establishing processes to respond to copyright claims.
→ Organizations are also exploring technical solutions like digital watermarking and content recognition systems to identify protected works. The key is balancing innovation with respect for creators' rights.
Other options could be :
→ Opt outs:
Organizations that don't want their copyrighted content used to train AI models can use opt-out processes. For example, OpenAI allows website owners to block AI systems from scraping their content. This gives content owners control over how their material is used while providing AI companies a clear way to respect these preferences.
→ Liability considerations:
Companies may hesitate to use AI tools due to copyright concerns. Without clear legal guidance on AI-generated content, businesses worry they could face lawsuits for unintentionally using copyrighted material. This uncertainty around legal liability makes some commercial users cautious about adopting generative AI services.
→ Explore technical guardrails:
Companies can implement technical safeguards to avoid copyright issues with AI tools. For example, Microsoft's Copilot uses content filters, monitoring systems, and abuse detection to prevent generating copyrighted content. These technical solutions help organizations use AI responsibly while protecting creators' rights.
→ Generative AI requirements:
The European Union is setting clear rules for AI copyright compliance through 2 key requirements to protect intellectual property rights while allowing AI development to continue responsibly :
- Transparency: AI companies must publish detailed documentation about what content they used to train their models, including copyrighted materials.
- Copyright Policy: Under the EU AI Act, AI providers must create and follow specific policies that identify and respect copyright laws.
To Conclude
The ever-changing landscape of copyright law faces unprecedented challenges with the rise of AI technology. Traditional copyright protection, which automatically safeguards creative works in most countries, now extends to both physical and digital creations. Copyright holders have economic rights to profit from their work and moral rights to protect their reputation and attribution. While fair use allows limited use of copyrighted material without permission, courts carefully evaluate factors like purpose, nature, amount used and market impact to determine legitimacy.
Major lawsuits between content creators and AI companies highlight current tensions, particularly regarding the use of copyrighted materials for AI training. Companies like Microsoft are implementing protective measures, including assuming liability for commercial customers and using technical safeguards like content filters. Organizations concerned about their content being used for AI training can opt out through established processes.
The European Union leads in regulation, requiring AI providers to be transparent about training data and implement formal copyright policies. As this field continues to evolve, organizations are developing various strategies to balance innovation with intellectual property rights protection.
Interested in learning more about AI Privacy and Data Protection ? Read our blog articles and contact us to help you build your AI Governance strategy.
Source : IAPP AI Governance in Practice Report 2024.