Summary
The Rise of AI Disclaimers
- New Clauses: AI’s unpredictable nature (and tendency to “hallucinate” errors) has led to new AI-as-a-Service (AIaaS) contracts and specific risk-allocation clauses.
- Shifting Risk: Suppliers use disclaimers to shift the responsibility of verifying AI outputs onto the customer.
Common Disclaimer Clauses
- Suppliers do not guarantee: That the AI is error-free, fits specific customer needs, or produces unique/accurate content.
- Customers must agree: That outputs are not professional advice, they are solely responsible for verifying data, and they won’t blame the supplier for faulty input data.
B2B vs. B2C Legal Impact
- B2B (Business-to-Business): Disclaimers can limit liability or create “contractual estoppel” (preventing customers from suing over known risks). However, clauses must be highly specific to pass the Unfair Contract Terms Act 1977 “reasonableness test.”
- B2C (Business-to-Consumer): Disclaimers are much harder to enforce because the Consumer Rights Act 2015 gives consumers statutory rights (quality, fitness for purpose) that contracts cannot legally override.
Takeaway
- Precise drafting is essential. Specific disclaimers work well to transfer risk in B2B contracts, but B2C contracts require extreme care to be legally viable.
Introduction
The adoption of AI is becoming increasingly popular for businesses across the UK and EU, and from it has emerged new types of contracts and clauses governing the commercial use of AI models.
The AI-as-a-Service (“AIaaS”) agreement is a prominent example, along with bespoke AI clauses within traditional contracts which are specific to the AI industry.
As AI adoption grows, so too does concern about legal liability. Suppliers seek to limit responsibility for AI-generated errors, while customers want assurance that AI tools are reliable, lawful and fit for purpose. The result is a growing trend towards the inclusion of AI-specific disclaimers in commercial contracts.
However, simply inserting a broad disclaimer stating that “AI outputs may be inaccurate” is unlikely to provide complete legal protection under English law. Businesses need to understand both the benefits and limitations of contractual disclaimers and ensure that AI-related risk allocation is carefully considered.
This blog examines the role of AI disclaimers in UK contracts, the key legal considerations under English law, and practical drafting points for both suppliers and customers.
Why Are AI Disclaimers Becoming Common?
AI systems operate fundamentally differently from traditional software.
Conventional software is generally designed to produce predictable outputs when given specific inputs. AI systems, by contrast, generate probabilistic responses based on patterns identified during training. Even sophisticated AI models can produce incorrect, misleading or fabricated information (a phenomenon often referred to as “hallucination”).
For suppliers, this creates significant risk. A customer may rely on an AI-generated output when making a commercial decision, only to discover later that the information was inaccurate or otherwise skewed.
Examples include:
- AI-generated legal or regulatory guidance that is incorrect.
- AI code generators generating flawed code or code with security vulnerabilities.
- Recruitment tools producing biased recommendations.
- AI-generated reports containing fabricated sources or data.
- Customer service chatbots providing inaccurate advice.
Given these risks, suppliers increasingly seek to include contractual language making clear that AI outputs should not be treated as definitive advice and that customers remain responsible for reviewing and validating results.
Examples of AI Disclaimers
An AI disclaimer is a contractual provision intended to limit expectations regarding AI-generated outputs and allocate responsibility between the parties.
Common examples include statements whereby ‘the Supplier does not warrant’ any of the following:
The use of the AI software will be uninterrupted or error-free;
No piece of software is perfect, and there are pre-conceptions around AI models more than any others that they will operate seamlessly – you will want to notify your customers that this is not the case ahead of any complaints or claims that your faulty software constitutes a breach of contract.
The AI software itself or its outputs will meet the customer’s requirements or expectations;
Especially when distributing your AI model to many customers, it will be impossible to fully understand each of their needs and requirements or anticipate them during programming. This disclaimer notifies the customer of this, and allows them to proceed in full understanding.
The AI software has been trained for the specific purpose of its use by the customer;
Similarly to the previous point, this takes the weight off your shoulders during the programming phase – you can maintain your vision for your AI model, and you can leave it with the customer as to how they wish to use it.
Where the AI software’s source code has been generated itself by generative AI such as ChatGPT, that it was built on reliable and sufficient data;
It is incredibly difficult, if not impossible, to collate the data used to train generative AI models – this point explicitly notifies the customer of this fact, and that they should proceed with caution if necessary, while also omitting your own liability.
Any generated outputs are unique, complete or accurate;
You will notice this disclaimer on publicly available AI sites such as ChatGPT and Gemini – again, its purpose is to warn your customers that your model isn’t perfect, and that they should adjust their expectations. This is particularly relevant should the AI model generate output that infringes on other copyrighted works or hallucinates.
There may also be another disclaimer but instead aimed at the understanding of the consumer. This will typically include statements whereby ‘the Customer acknowledges and agrees’ to the following:
The AI software and the generated outputs are not fully accurate and do not constitute professional advice;
This disclaimer is specifically aimed at avoiding a professional negligence claim and discharges the relevant liability there.
The customer should review the accuracy, completeness and correctness of all generated outputs, and is solely responsible for this;
This places the onus for the verification of the generated output data onto the customer themselves, thereby removing any liability from the supplier should that data be false and cause the customer to incur financial losses.
They will not represent to any third party that the generated output was verified by the Supplier.
Again, this places the onus regarding the accuracy of the generated output onto the customer and lifts it away from the Supplier. It also prevents the Supplier from being dragged into any disputes that occur between the customer and a third party arising from the customer’s use of the AI system.
That the supplier has no liability for loss attributable to the customer’s data;
Many generative AI solutions rely on customer data (input) to generate output. If the input is defective, it is unlikely to produce an accurate or useful result.
The Customer’s Perspective
For obvious reasons, customers should approach AI disclaimers with caution.
Whilst suppliers will often seek to exclude liability for inaccurate outputs, system errors or other AI-related shortcomings, customers should be careful not to accept contractual provisions that transfer an unreasonable amount of risk onto them.
The starting point for any customer should be to assess the intended use of the AI system. The more critical the AI tool is to the customer’s business operations, the more difficult it may be to accept broad disclaimers. For example, a customer using AI to assist with drafting internal documents may be willing to assume responsibility for reviewing outputs. By contrast, a customer relying on AI to support regulatory compliance, financial decision-making, recruitment or healthcare-related functions will likely require stronger contractual protections.
Customers should also take the profile of the supplier into account when thinking about whether or not to accept an AI disclaimer. Suppliers that hold themselves out as offering a tool that works wonders in a niche area should, arguably, be giving stronger protections than a supplier who is offering a solution that fits a variety of use cases.
Effect of AI Disclaimers
The effect of an AI disclaimer depends on if the contract is a B2B or B2C contract.
B2B
In B2B contracts, the intention with a disclaimer is to limit or totally exclude the covered liability. Mostly, but not always, the party using a disclaimer is to limit or exclude liability for negligence – that is, a failure to perform services to a particular standard.
Disclaimers are not always absolute and suppliers can still fall below the standard that is being disclaimed.
For example, a disclaimer that a software is not guaranteed to be free from viruses does not always mean that a horrifically damaging and easily avoidable virus uploaded onto a customer’s system would not be actionable by the customer.
This is for two main reasons:
- a general disclaimer may not be specific enough to cover the liability; or
- the disclaimer fails to satisfy the “reasonableness test” under the Unfair Contracts Terms Act 1977, although specific disclaimers that are properly drafted are less exposed to challenge than general unconsidered disclaimers.
Disclaimers can also operate as a form of contractual estoppel, whereby one party agreeing that a state of affairs is true is precluded from arguing that the start of affairs is not true. To use an example, if a customer acknowledges that not all AI output will be accurate, it could be prevented from bringing a claim on the basis that it received inaccurate outputs from the supplier’s AI system.
B2C
With consumers, however, disclaimers will be much harder to rely on at all.
B2C contracts are subject to the Consumer Rights Act 2015 (CRA). Under the CRA,
This is because the provisions of the Consumer Rights Act 2015 will apply. Under this Act, individual consumers have statutory rights that set a standard that a supplier cannot go under in the provision of goods or services, including that digital content (such as software) must be:
- of satisfactory quality
- fit for the purpose that the digital content was sold for and
- sold as described
Because these statutory limits cannot be modified by contract, an AI disclaimer that seeks to modify them could be challenged by the consumer.
However, it is worth noting that the statutory terms are not absolute and will vary depending on the context. For instance, a glitch in a video game that does not materially affect the enjoyment or functionality of the game will not comprise a breach of these rights, whereas a car which was delivered with a missing wheel, thereby making it undriveable, would constitute a breach. AI systems will likewise vary in context and purpose.
Conclusion
It is therefore highly recommended that you include AI disclaimers in your B2B contracts. They can help remove all liability in the event that your AI model falls short of the desired quality, and transfer this liability onto the customer. It is imperative, however, that you are as precise and specific as possible in these clauses to ensure that they are deemed legally effective.
In the case of B2C contracts, however, the precise drafting must be considered with great care if they are to be effective.
EM Law are experts in AI law and AI contracts. If you need any help with the issues referenced above, please do not hesitate to get in touch with us here or visit our AI Lawyers, Software & Tech Lawyers, and Contract Lawyers pages for more information.
