Why should investors monitor IP rights?

• Intellectual property rights (IPRs) claims can result in court orders requiring a company to immediately stop using the contested IP. If a company can't find an alternative way to operate without the IP, it might have to abandon the product entirely, posing a substantial financial risk.
• IPR claims can also result in damages, which can substantially increase the costs of an AI project, potentially making it financially unviable.
• While rare, some IP violations can lead to criminal charges.

What IP rights are relevant?

IPRs most relevant for AI include copyright (protecting “creative works”, e.g. literature, source code), trademarks (protecting brand identifiers, e.g. logos, product names); and the right of publicity / right in one’s own image (defending against unauthorised commercial use of an individual’s likeness, e.g. face, voice), and patents (protecting inventions).

How can AI models conflict with IP rights?

During the input stage, AI system developers and deployers can face IP rights issues when they use protected materials to train, fine-tune or feed AI models. Collecting licenses to avoid conflicts may make AI project unsustainable.
During output stage, AI developers might face IP issues if the AI creates content too similar to copyrighted works, such as artwork resembling a protected painting. The unpredictability of AI outputs, due to its “black box” nature, complicates monitoring and correcting potential infringements.

Why do IPR owners bring claims against AI companies?

Content creators and celebrities for revenue protection and artistic control, publishers for revenue protection; corporations for brand protection; individuals for privacy, control over personal data; academic institutions for research recognition, IP security, revenue protection.

Is the legal situation clear?

No, numerous court cases are pending trying to answer fundamental questions of law–mainly in the US and the U.K.

Which AI companies are most exposed to IRP disputes?

Generative AI models, synthetic media/“deepfakes”, and facial recognition providers.

Which IPR is the biggest challenge currently?

Copyright, due to its many pending cases and vocal concerns from writers, artists, and musicians about AI diluting their markets.

What are the key defences for AI companies?

• The primary defence for AI companies is fair use, which permits limited use of copyrighted material for specific purposes like commentary, search engines, criticism, parody, news reporting, research or scholarship.

• Another defence is that AI systems merely extracts unprotectable ideas and patterns from data, which does not equate to copying, redistribution or recommunicating of the works themselves.
• AI companies may claim inputs are not copyrightable if they involve non-original, trivial, or widely known data, or are based on public domain government works.

What can AI companies do to mitigate IPR conflicts?

• AI companies can avoid using IPR-protected materials, though determining what is genuinely open-source can be complex and costly.
• AI companies can seek licenses from rights owners, but this approach can be challenging for developers of LLMs due to the high costs and complexities involved in identifying and negotiating with numerous rights holders.
• AI companies can use media liability, cyber, and general liability insurance to mitigate risks, though they must carefully review policies for exclusions related to intentional acts and violations that may not cover all AI activities.