AI x Dispute Resolution: AI and IP - protecting rights & managing risk

From landmark copyright cases to ownership questions and cross-border risk, this quick guide distils the latest legal developments and the practical steps organisations should take now.

AI-related disputes are no longer a hypothetical risk for organisations rolling-out use of AI models and tools. With cases already before courts across the UK, Europe and the US, early rulings are beginning to reshape the risk landscape, particularly in terms of IP infringement.

In the latest session of our AI x Dispute Resolution Webinar Series 2026, Joel Smith, Sarah Bailey and Christina Thompson examined the key cases on AI training and outputs, whether AI-generated inventions and outputs are capable of IP protection and set out the practical steps organisations should be taking now to manage risk. This article summarises the key issues discussed, the practical takeaways, some of the interesting questions raised by attendees and what organisations should be thinking about next.

AI and IP flashpoints to watch

• Getty v Stability AI: a landmark UK trial, but no clear answer

Getty alleged that Stability AI infringed both copyright and trade marks, through training of its model and outputs, but Getty’s main copyright claims failed and it succeeded only narrowly on trade mark infringement. An appeal is to be heard starting 10 November 2026. The UK Government’s proposal to change the law and permit text and data mining for commercial use was abandoned earlier this year.

• Germany draws a harder line on AI training

The Munich courts have taken a stricter approach, finding in GEMA v OpenAI that ChatGPT’s reproduction of protected lyrics amounted to copyright infringement and holding OpenAI responsible for the outputs. GEMA v Suno extended this reasoning to AI-generated music, signalling a high infringement risk where protected works are used and there is memorisation in training the model.

• Who owns AI-generated ideas and content?

To seek patent protection, inventors in the UK, US and Europe must be human, as confirmed by the DABUS cases, although the level of human contribution required for AI-assisted inventions remains unclear. Organisations should also avoid entering details of unfiled inventions into open AI systems, as this could amount to a public disclosure and undermine patentability.

Copyright protection generally depends on demonstrating sufficient human creative input and control in generating outputs. The UK Government has proposed removing the specific protection for computer-generated works. AI cannot be a joint author, and platform terms may also determine ownership.

Six actions businesses should take now

• Check provider training practices. Confirm data sources and licensing of data, address these in contractual protections.
• Test for memorisation. Spot-check outputs and use safeguards such as filters and guardrails.
• Do not rely on text and data mining exceptions. Their protection may be narrower than expected.
• Train employees. Set clear rules for AI inputs, outputs and escalation of concerns.
• Record human contribution. Keep prompt logs and other evidence supporting IP protection and ownership.
• Assess risk by jurisdiction. Copyright and AI rules differ across borders.

The questions keeping businesses awake

The session generated a lively Q&A, and the speakers picked out a few themes in their closing remarks. Several questions from the floor also stood out:

• Will trade secret protection replace reliance on copyright? The speakers thought this was likely as purely AI-generated works may lack sufficient human creativity for copyright protection.
• Will major AI providers offer ownership warranties? Their strong bargaining position may make meaningful warranties or assurances on provenance difficult for businesses to negotiate and secure.
• How should businesses manage fixed vendor terms? Where vendors offer non-negotiable terms, organisations may need to rely more heavily on internal controls than contractual protection.
• Has UK copyright authorship been tested? Unlike patent inventorship under the DABUS cases, this remains unresolved and it will be important to see if a case on AI-generated works reaches the courts.
• Why does technical evidence vary? All of the cases mentioned involved the submission of extensive technical evidence, from experts and scientific papers. However, the position on memorisation in training will vary by model, including how it was trained and how it functions.

What comes next: more scrutiny, more disclosure, more risk

How AI systems are trained and used is coming under closer scrutiny, with increased consequences for non-compliance. Disputes between AI developers and rights holders are increasing, particularly in the US, but the courts are taking different approaches to similar issues. For now, outcomes will continue to depend heavily on the facts of each case and the jurisdiction in which it is heard. At the same time, the EU AI Act’s enforcement powers have provoked increased scrutiny of training data summaries and copyright policies and more transparency. The market is responding through more one-off licensing deals and emerging collective licensing options.

Organisations should expect increased audit and disclosure requests from rights holders seeking information about the models. They should therefore ensure their approach to governance and use of AI is robust enough to withstand increasing scrutiny and potential enforcement.

If you would like to discuss any of the issues raised in this article, please contact any of the speakers linked above.