I. Intellectual Property Issues in AI

A. IP infringement

1. Copyright infringement—creative works

Imitation is the sincerest form of flattery? Not so much when it’s generative AI, which trains on voluminous data and creates new works. It inevitably plagiarizes excerpts of books and articles and copies elements of art, photography, and music compositions with impunity.

From an artist’s perspective, not only are your creative works being stolen, but your livelihood is jeopardized by AI. The recent Hollywood writers and actors strikes reflect nothing less—click here.

2. Copyright infringement—software coding

There’s a specific wrinkle for copyright issues when generative AI creates software code. The software community has a unique and proud tradition of crowdsourcing for open source software—click here. Contrary to popular misunderstanding, open source software is not by definition “free.” Instead it can and often is incorporated into closed (i.e., proprietary) software, for commercial sale.

But one particular type of open source software, cheekily named “copyleft” software, is an exception. Copyleft licensing terms are described as “viral,” because they require all derivative works incorporating copyleft code to also be released under a copyleft license.

This gives rise to a possible scenario where generative AI incorporates excerpts of copyleft open source software as part of its plagiarizing process only to have the software that it generates rendered unprotectable because of it. While most commentators do not believe that a court would ultimately impose such a harsh result should a dispute arise, it is a position available to any litigant who would benefit from it. Software businesses must take note of this risk and take steps to mitigate against it.

3. Right of publicity misappropriation

There’s presumably a special place in hell for whoever knocked off Tom Hanks, right—click here? And yet, the internet goes far lower in misappropriating the images and likenesses of famous and every day people alike.

Any generative AI’s creation of a human image is trained on the images of one or more people. Virtually none of those people have granted permission for the use of their images.

There is no federal law in place granting any “right of publicity” preventing the unauthorized commercial use of an individual’s name, likeness, or other recognizable aspects of one’s persona. The right of publicity is but a patchwork of state and common law.

Every state has its own flavor of right of publicity law, with some states’ laws being more established than others. My home state of Arizona, for example, has two statutes recognizing a right of publicity, but directed only at soldiers. In 2020, an Arizona superior court held this reflected a legislative intent to deny a right of publicity for civilians. This was overturned on appeal, with the court holding that Arizona always has and continues to recognize a common law right of publicity.⁴

A typical AI provider’s business model is to sell software-as-a-service (SaaS) subscriptions for the use of its platform over the cloud. The provider is thus subject to both federal law and the individual laws of every state.

B. Protection of IP rights by AI providers

The global wave of legislation on AI has focused on mitigating against its potential negative effects on society. Not much specific has been directed toward defining the IP rights of providers to date.⁵

1. Authorship / inventorship issues

Will the law recognize generative AI as capable of creating original works or inventions for IP ownership purposes? And will individuals with ownership rights over such output be granted copyright or patent rights?

Some argue that we should treat AI like a tool. And creators should be able to use it like any other tool to create original works or inventions and secure IP rights on them.

The U.S. federal courts appears to have set an outer boundary to this position over the past few years. Two courts have held that if there is no human hand in any part of the generative AI’s process, then neither copyright rights nor patent rights are available to the AI’s owner.⁶ But the question of whether creative works or inventions made under the direction of human beings and with only some degree of non-human assistance should be eligible for copyright or patent protections remains wide open.

2. Trade secret protections

Many AI companies rely primarily if not exclusively on trade secret protections to protect their innovations. As noted in my blog last week, AI seems like a great candidate for IP protection by trade secrets.

But President Biden’s Oct. 2022 Blueprint for an AI Bill of Rights and Oct. 2023 Executive Order on AI make clear our current administration’s intent to make it a heavily regulated industry. Both the EU’s proposed June 2023 and China’s Aug. 2023 versions do the same. Providers and implementers may need to meet future regulatory standards establishing the accuracy and lack of “algorithmic discrimination” of their AI—click here.

When an AI company’s invaluable trade secret rights collide with the government’s interests in protecting its citizens in this highly disruptive area, who do you think will win? And who and what should we be rooting for? Let’s play this out in next week’s blog….

II. Legal Risks For Businesses and AI

The business use of AI gives rise to various types of legal liabilities, including:

• IP infringement,
• “algorithmic discrimination,”
• data privacy violations, and
• fraudulent and negligent misrepresentation.

In the U.S., your business will be subject to greater regulatory scrutiny if it involves a “sensitive domain.”⁷ These include:

• hiring,
• financial systems,
• healthcare,
• housing, and
• education.

Your primary defense against an AI claim will be fundamentally an issue of compliance, with:

• existing principles of IP and technology law,
• existing applicable regulatory regimes, and most importantly
• the AI-specific legal and regulatory regimes that various national and state governments will develop in the coming months and years.

A. AI use in your hiring, promotion, and termination processes

All businesses implementing AI into their hiring, promotion, and firing processes will need to implement a defensible AI. The employer and/or its AI provider will be liable for any “algorithmic discrimination” harming an affected claimant. If a jilted individual files any employment discrimination or unlawful termination suit and the implementation played any sort of part in the process, you will need to defend it.

B. Incorporating AI output in your service offerings

When implementing AI in providing services to your customers, then you and/or the provider face the following legal risks:

1. IP infringement (for overview of legal risks, see Sect. I above)

2. Data privacy violations

There is an insatiable need for the collection of data in AI for the training of its models. Each data set will, especially if unchecked, invariably contain the personal identifying information (PII) of individuals. And some AI implementations (e.g., for surveillance or for targeted marketing) collect and intentionally generate PII.

Each data set thus becomes a potential target for a data breach. And if a breacher makes this information public, for e.g., by posting on the Internet, then such PII repositories again become pulled into data sets for the training of other AI models, continuing the cycle.

Such collection of PII gives rise to potential data privacy liability for the AI implementer or provider, e.g., for:

1. improper collection and use of PII,
2. improper disclosure of the collection and use of PII,
3. lack of proper cybersecurity against data breaches, and
4. improper transfer of data containing PII.

Data privacy concerns further arise from an AI implementer or end-user uses the AI platform. The information that the user enters as prompts for the AI to generate a desired response may contain PII and itself be collected and used by this or other AIs in the future.

A veritable alphabet soup of federal and state laws may apply depending on the situation—click here.

3. Fraudulent and negligent misrepresentation

Fraud in any form, including that intentionally perpetrated through AI, gives rise to criminal and civil liabilities. The only open question is what additional laws and regulations will come into effect specific to AI.

An “AI hallucination” is content generated that is nonsensical or unfaithful to the provided source content. They happen with surprising frequency and, if and when relied upon, can cause harm. The liability to be borne by implementers or providers for such hallucinations is yet to be determined.

C. Addressing unauthorized use of AI by your employees

Have you updated your employee handbook for AI issues?

All companies should update their company policies and employee handbooks to address any unauthorized use of AI by their employees. At a minimum, you should require your employees to disclose any such use made in the course of their employment.

This will help mitigate against the risks of any of the types of third-party claims (IP infringement, algorithmic discrimination, etc.). It will further mitigate against claims of unlawful termination for such conduct.

D. Will there be federal protections for providers?

Section 230 of the Communications Decency Act of 1996 has long shielded internet platforms from liability for content created by users. Title II (the “Online Copyright Infringement Liability Limitation Act”) of the Digital Millenium Copyright Act of 1998 limits the liability of online service providers for copyright infringement. Some credit such federal protections as being instrumental to the very success of the internet.

Will our government apply these same principles to protect providers from liability for content generated using their platforms? Should we want them to? Stay tuned.

E. Indemnification

The indemnification terms in technology contracts delineating which party is responsible for defending against third-party claims (e.g., for IP infringement) are typically heavily negotiated. This has and will continue to be the case for contract negotiations between AI providers and implementers.

Will our government continue to allow parties the freedom to contract on indemnification terms, e.g., for intellectual property issues? Or will AI be the subject of more oversight on this and other issues?

The topic of indemnification has had a special place in my heart for decades. It is the zero-sum game of zero-sum games. More to come on this.