Summary
- Importance: AI clauses are becoming increasingly important as businesses adopt AI tools and services. They help allocate risk and provide clarity on issues that traditional contracts often do not address.
- Key areas covered by AI clauses: AI-generated errors, intellectual property ownership, data protection, confidentiality, bias and discrimination risks, regulatory compliance, and liability for AI-assisted decision-making.
- Solutions:
- AI clauses can address commercial issues such as token-based pricing, usage limits and other AI-specific charging models.
- Contracts should clearly define what constitutes an AI system or feature to avoid disputes and ensure liability and disclaimer provisions are appropriately targeted.
- Intellectual property provisions should clarify ownership of AI outputs, rights to use customer data for model training, and the scope of any IP infringement indemnities.
- Liability clauses should deal with inaccurate or misleading outputs, the allocation of responsibility for decisions based on AI recommendations, and the risks associated with autonomous AI agents.
Introduction
AI clauses are becoming an increasingly important feature of modern commercial contracts as AI becomes embedded in everyday business operations.
From AI-powered customer service tools and document review software to generative AI platforms such as ChatGPT and Microsoft Copilot, organisations are increasingly relying on AI to improve efficiency and reduce costs. AI developers and providers are likewise experiencing high customer demand and are entering into customer contracts at scale.
Traditional contracts often fail to address the unique legal and commercial issues that arise when AI systems are involved. As a result, businesses are increasingly turning to AI clauses to allocate risk, establish responsibilities and provide certainty between contracting parties.
In this blog, we explain what AI clauses are, why they matter and the key provisions businesses should consider when negotiations contracts involving AI systems.
What are AI clauses?
AI clauses are contractual provisions that specifically address the use, development, deployment or integration of artificial intelligence. AI clauses would cover a wide variety of different issues and risks, including:
- AI-generated errors or inaccuracies;
- Intellectual property ownership of AI outputs;
- Confidentiality concerns;
- Data protection compliance;
- Bias and/or discrimination risks;
- Regulatory compliance; and
- Liability arising from AI-assisted decision-making.
Key AI clauses to consider
Defining ‘AI’
Where an AI system is offered, these platforms typically consist of a combination of traditional “deterministic” software and AI-powered features. For example, a platform that may help businesses collate and respond to customer messages could also include a feature that auto-generates a potential response using a generative AI model like ChatGPT.
Without a clear definition, disputes can arise over whether particular software, tools or functionalities fall within the scope of the agreement and what the liabilities are depending on what functionality is used.
Take a scenario in which AI features can be toggled off and on within a platform. It would not be commercial for a supplier to disclaim liability for the accuracy of the platform as a whole, where they should be doing so only for the AI feature that has been enabled. If ‘AI’ is not properly defined, the supplier is at risk of casting such disclaimers too widely.
This can have commercial implications: customers are unlikely to accept broad and untargeted disclaimers. In England, it is possible that lumping all functionalities of a platform together and applying AI appropriate (but inappropriate for traditional software) disclaimers in standard terms and conditions may leave the agreement open to challenge under the Unfair Contract Terms Act 1977(UCTA).
UCTA states that a party is unable to fully restrict liability unless doing so would be “reasonable.” It is likely that widely casting liability protections, without the need to do so, is going to be much less reasonable than targeted, AI specific disclaimers in a contract.
Using AI clauses to specifically define what is meant by an AI system, tool or feature can also assist with clarifying the intellectual property position.
To use an example: a properly drafted software contract should state that all intellectual property rights in the platform remain with the supplier. By defining what an AI is in detail, it will be clear that all components of an AI system remain with the supplier (for example, a AI model’s variables and weights that may change in response to customer input).
Commercial aspects
Most AI systems require a significant degree of computing power that can come at a high cost to the supplier (or indeed the customer!).
The data that an AI processes is broken down in pieces, called “tokens.” In the English language, 1 token is usually about 4 characters. Tokenisation enables the computational requirements (e.g. electricity or time) to be priced with precision – usually a single dollar figure per million tokens.
That said, AI systems have a wide variety of pricing models.
AIaaS systems are frequently charged at a flat monthly or annual fee per customer. If one customer uses an AI system much more than the rest, that can make the flat monthly fee look less attractive to the supplier who is having to incur token costs at a higher rate than other customers.
Conversely, if token costs are passed through to the customer on a usage based model, then it will be important to clearly define what counts as usage so there is no overpayment on the part of the customer or loss on the part of the supplier.
AI systems can also be a mix of commercial approaches. It is not uncommon to see “tiered” plans with different usage limits for different flat fees.
AI clauses can assist businesses of all sizes in dealing with these issues that are both specific to AI and also differ slightly to most traditional approaches to software licencing.
Intellectual property
We have touched on above how clearly defining AI can assist in clarifying the ownership of intellectual property connected with AI systems.
AI clauses can also help deal with other related issues specific to AI when it comes to the licencing and/or ownership of intellectual property. These include:
- Ownership of AI outputs. For AI systems that are generative in nature (e.g. large language models that can generate text), it can be unclear who owns the output that is generated. In the UK, AI generated output can be protected by copyright – but whether the original developer or the user who entered into the prompt owns it automatically is debateable. To deal with the ambiguity, it is important to clarify in an agreement for an AI platform who owns the output – the customer or the supplier?
- Data usage and training rights: AI systems can be improved by being given access to a wide variety of data. A common contractual issue is whether the supplier can use new customer data provided for this purpose. Customers will typically be hesitant to allow this, as the way in which an AI can “learn” from data can expose the customer’s data to third parties. Any personal data that is used for the improvement of an AI can also fall out of the control of the original controller; that is unlikely to comply with the UK GDPR. But suppliers can approach this issue in a more targeted way – obtaining ownership of limited customer generated datasets that can be anonymised before the improvement process begins.
- Intellectual property indemnity: it is standard for software suppliers to give an indemnity that their platform (and the customer’s use of that platform) will not infringe the intellectual property rights of any third party. However, as the ownership of AI outputs can be ambiguous, it is common to see AI outputs being excluded wholly or partially from intellectual property indemnities on the basis that the supplier does not exercise full control as to how they are generated or whether that generation may infringe someone else’s property right. A common negotiated position is that the indemnity only extends to an intellectual property claim which is not attributable to the customer’s data.
Liability
AI systems pose novel liability challenges which are crucial for both suppliers and customers to deal with. AI clauses are necessary to manage these risks. Specific concerns include:
- Accuracy and performance disclaimers: because AI systems are fundamentally predictive in nature, they can produce outputs that are inaccurate, misleading or entirely incorrect (often called “hallucinations”). It will be important for suppliers to make sure that their contracts specifically reflect this risk and disclaim liability for inaccuracies. Customers, on the other hand, may wish to set limits to this or put in place reliability or quality controls that enable the customer to obtain recourse if the AI system is consistently producing inaccuracies.
- Decision making allocation: AI tools can also make predictions or recommendations. They can also evaluate performance and other metrics. Some AI systems can do this automatically without human input – placing liability for the decision on the supplier. As such, it will typically be appropriate to allocate responsibility for decision making. Most suppliers insist that any AI recommendation should be subject to human review and decision making.
- Agentic systems: going further, some AI tools offer “AI agents” – computer programs that can interact with other systems and make decisions with varying degrees of autonomy. Responsibility for the actions of AI agents must be carefully considered and allocated using contractual mechanisms.
Regulatory compliance
Unlike traditional software, AI systems can raise a wide range of legal and regulatory issues, particularly where they process personal data, influence decision-making or interact directly with consumers.
As a result, contracts should clearly allocate responsibility for regulatory compliance between the parties. Much of the liability will rest on the supplier, but customers can influence whether the use of an AI system complies with applicable laws.
For example, a supplier may be responsible for ensuring that the AI technology itself complies with applicable legal requirements, while the customer may be responsible for ensuring that the AI is used lawfully and appropriately within its organisation.
Many businesses are also seeking contractual warranties that AI systems have been developed, trained and tested in accordance with applicable laws and industry standards. Customers may request assurances that the AI has been subject to appropriate governance procedures, bias testing, security assessments and human oversight measures before deployment. These responsibilities can also be shared. In the UK public sector, government organisations are subject to an equality duty which requires them to ensure that their use of AI systems is not discriminatory.
In some cases, organisations may also seek AI clauses that contain audit rights allowing them to verify compliance with legal and regulatory obligations. These provisions can be particularly important where AI is used in highly regulated sectors such as financial services, healthcare, insurance, recruitment or public sector procurement.
AI specific regulation is also being developed, such as the EU AI Act. AI clauses can also be used to ensure that were regulation changes, the parties have flexibility to comply with new requirements as they emerge.
AI clauses – key takeaways
Well-drafted AI clauses can provide clarity, allocate risk appropriately and help businesses use AI technologies with greater confidence. Whether you are procuring AI-powered services, licensing AI software or developing AI solutions, reviewing your contracts to ensure they contain suitable AI clauses should now be a key part of your risk management strategy.
At EM Law, we are experts in AI law. If you need help navigating AI, AI clauses or any other issue, please contact us here or visit our AI Lawyers or Software & Tech Lawyers pages for more information.
