Although one may think that artificial intelligence (AI) and copyright are issues separately addressed, nothing could be further from the truth, particularly today. With the rise in popularity of tools such as OpenAI’s ChatGPT and Stability AI’s Stable Diffusion, EU policymakers decided to regulate the use of generative AI (GenAI) systems, i.e. the technology at the intersection of AI and copyright. For the startup ecosystem, generative AI systems bring immense opportunities: they are a new means of scaling, thus allowing startups to compete with more established players, in sectors ranging from creativity to health.

Unfortunately, instead of focusing on these opportunities for smaller players, the European Union’s AI Act focuses on the risks. By covering general purpose AI (GPAI) models in its text (therefore including GenAI), EU institutions have also introduced provisions that overlap with the EU’s Copyright Directive, thus creating more complexity and challenges than opportunities for startups developing GPAI models.

What do the AI Act and the Copyright Directive say?

In dissecting the specifics of the AI Act and the Copyright Directive, it’s imperative to grasp the nuanced provisions that directly impact startups operating at the confluence of AI and copyright regulations.

The AI Act indeed provides that GPAI model providers must:

• Establish a policy aligning with Union copyright laws. This involves respecting reservations of rights under Article 4(3) of the Copyright Directive. The integration of state-of-the-art technologies is encouraged, although the exact purview of these technologies awaits clarification in upcoming guidance. Importantly, the Copyright Directive’s Article 4 empowers rights holders to reserve their rights concerning text and data mining (TDM).
• Draft and publicly share a sufficiently detailed summary of the content used in training. Following a template provided by the AI Office, this summary aims for comprehensiveness rather than technical minutiae, as explained by accompanying Recital 60k. Notably, the AI Act acknowledges the necessity of tailoring obligations to the provider’s size and contemplates simplified compliance for SMEs, including startups.

Turning our attention to the Copyright Directive, Article 4 delves into the pivotal realm of text and data mining. It introduces exceptions to copyright for:

• TDM for research purposes: This exception in Article 3(1) allows reproductions and extractions of lawfully accessed works and subject matter for scientific research purposes. However, its applicability is confined to research organisations and cultural heritage institutions.

• TDM beyond research purposes: Article 4 establishes a distinct copyright exception for text and data mining beyond scientific research, provided users have lawful access. Crucially, rights holders retain the option to opt out by expressly reserving their rights.

The AI Act explicitly references Article 4(3) of the Copyright Directive, emphasising that any use of copyright-protected content necessitates authorisation unless pertinent exceptions apply. Specifically, providers of general-purpose AI models must seek authorisation from rights holders for text and data mining on works where the right to opt out has been expressly reserved.

This intricate interplay between the AI Act and the Copyright Directive underscores the complex regulatory landscape startups must navigate. The challenge lies in finding the delicate equilibrium between fostering innovation and ensuring compliance within the European Union’s evolving AI and copyright framework. 

What are the key takeaways?

To sum it up for startups diving into the world of AI and copyright rules, here are a few key things to keep in mind:

1. Get materials legally: Make sure you’re getting your hands on copyrighted materials the right way. Whether you’re buying them, getting a licence, or using publicly available stuff, make sure it’s all legal and above board.
2. Watch out for opt-outs: Some folks who own the rights to stuff can say “no” to letting you use it for Text and Data Mining (TDM). So, check if they’ve put a stop to TDM, and plan accordingly.
3. Follow the rules of your licence: If you’re using copyrighted materials through a subscription or licence, pay close attention to what the rules say about TDM. Stick to these rules carefully to avoid any legal hiccups.
4. Keep things secure: Any copies of works you make for TDM need to be kept safe and sound. Don’t share them illegally. Make sure your data security measures are up to snuff.