In July 2026, the UK Jurisdiction Taskforce released a paper (“Legal Statement on Liability for AI Harms”) setting out how persons may be liable for harms caused by the (mis)use of AI (UKJT Paper), observing that the English common law’s approach to negligence is well suited to allocating the liabilities that will arise from a professional’s use of AI. For insurers, this clarifies that AI exposures may be implicit in existing financial lines policies (absent express exclusions).
Professional negligence for AI risks
Insurance cover for AI risks – to cover or not to cover?
Considerations for underwriters and claims handlers
Professional negligence for AI risks
AI has not been assigned its own legal personality (under the common law or statute), and nor does this appear likely from the UKJT Paper which considers that firms remain liable for an employee’s negligent AI use in the same way they would for any other negligent act.
English professional negligence law has developed incrementally to adapt to new professions, new ways of working and developing technology and the UKJT Paper considers that adapting to AI risks is no different. Some examples cited include:
- where a professional incorrectly uses an AI tool when it was inappropriate to do so, or fails to apply their own professional judgment;
- the wrong AI tool is used for the task at hand;
- the output of the AI tools is insufficiently checked for accuracy;
- confidential or privileged material is uploaded to an AI tool which has insufficient controls to protect the confidential nature of the material;
- failure to use AI when a reasonable professional would have used it; or
- from a causation perspective, where professional standards set an expectation that professionals maintain records for their decision making when using AI tools, but they fail to do so.
In addition to expert evidence, the development of industry guidance materials (particularly as those guides evolve from high-level to detailed as the use of AI becomes embedded and tested) will likely be relevant to how the Courts determine whether professionals have met their standard of care.
Insurance cover for AI risks – to cover or not to cover?
With the law of negligence already capable of assigning liability for harms involving the use of AI, this exposes Insurers to the risk that existing financial lines policies are already covering these risks even if that was not underwriters’ intention (unless excluded).
While always subject to policy terms, there can be implied or ‘silent’ cover in financial lines policies for AI related professional risks in the following ways.
Civil liability and professional indemnity policies
As professional indemnity policies are designed to cover claims for professional negligence, and the UKJT Paper clarifies that the law of negligence will respond to AI (mis)use cases, these policies appear a natural ‘fit’ for AI risks to be insured unless expressly excluded. Certainly, the examples above would appear broadly to fit within the boundaries of professional indemnity cover. Civil liability policies are equally capable of responding, particularly where they provide cover to ‘insured persons’. However, these policies may also respond to claims against insured entities. Take, for example:
- a bank that is the subject of a claim for damages where it is vicariously liable for its employees’ failure to supervise the outputs of AI tools, or because the bank adopted a tool for assessing mortgage applications when due diligence would have identified that the tool was unable to assess creditworthiness;
- a bank uses an AI agent to handle customer complaints in which the agent is held out to be a human, that agent incorrectly offers a monetary remedy and the bank refuses to honour the remedy once it is escalated to a human agent. A subsequent claim for payment could also fall for cover.
There will likely be complicated questions of contribution if the loss can be attributed to a systemic error in the AI tool that the developer ought to have known about.[1]
Contribution claims (at least initially) are likely to be costly to run until such time as they have been tested in the caselaw.
Where the professional negligence is widespread (e.g. in an accountant’s use of AI tools to prepare tax returns or an investment manager’s use of an AI tool leads to recommendations of prospective investments which do not fit the investment mandate) there may be the prospect of bringing such claims as a form of collective action, inevitably increasing the defence costs exposure. The use of AI potentially heightens the chances of multiple errors, mis-selling claims, and regulatory scrutiny over potential systems and control failings, which raise a number of coverage issues such as whether the regulatory proceedings fall within the boundary of the policy or whether the claims or losses aggregate.
Management liability policies
Management liability policies may respond where widespread misuse of AI tools by professionals in an organisation is attributed to a failure of management to exercise proper oversight. Not least, the oversight over the selection, implementation and use of AI tools, and the ongoing control frameworks for managing associated risks.
‘AI washing’ by professionals and financial services entities, seen in a number of actions in the United States involving corporates, could become a focus area for regulators. AI washing can involve entities overselling their use of, or the capabilities of their own developed AI tools which cannot be substantiated with evidence.
As the use of AI tools becomes more prevalent and the technology continues to advance, we can foresee regulators scrutinising whether AI marketing representations involve misrepresentations. Shareholder activists may also look to file shareholder claims under s90/90A FSMA or at common law (for misrepresentation, deceit or breach of contract) against listed entities on the basis that such representations have inflated share prices.
Other product lines
While the UKJT Paper covers professional negligence risks, Insurers should not lose sight of how other product lines could provide cover for ‘silent AI’ risks. For example:
- product liability policies, where losses can be attributable to systemic defects in an AI tool which the developer ought to have known about; and
- crime policies, where staff, contractors or external threat actors use AI tools (such as ‘deepfakes’) to defraud insureds.
Considerations for underwriters and claims handlers
Considerations for existing products and underwriting
The use of AI in professional and financial services firms is only going to increase, and while that will undoubtedly lead to an increase in notifications and claims, at least in the near term, that does not necessarily mean that widespread changes are needed to the way financial lines products are underwritten.
It will be incumbent upon underwriters to (if they haven’t already) incorporate AI risk assessment into their pricing considerations. Underwriters should be considering:
- how prospective insureds are using (and not using) AI in their business (and representing to the public how it is used) in its varied capacities (e.g. AI agents/chatbots, AI tools to provide or formulate advice etc.), and in particular whether riskier or ‘frontier’ models are used;
- what due diligence has been performed over the selection of AI tools to make sure that they are appropriate for the chosen tasks;
- whether there are limitations in supplier contracts that prevent the assignment of liabilities being passed up the chain;
- what training and guardrails have been put in place to internally govern the use of AI by staff and contractors;
- what AI risk control frameworks are in place to govern the monitoring of AI use, improvements needed in employee training and guidance, and to guide developments in or the adoption of new tools;
- where are insureds’ exposures to potential crime risks by staff or external threat actors (e.g. the use of deepfakes to perpetrate financial fraud);
- do insureds make representations at large on their use of AI and can those representations be supported by evidence (relevant to AI washing);
- whether existing policy wordings clearly capture, or exclude, AI-related conduct and harms;
- whether definitions (such as ‘professional services’ and ‘professional business’) might be wide enough to cover the use of AI if an autonomous model is used and the definitions refer to ‘natural person’;
- pricing, scope of cover and sub-limits for cover extensions, such as defamation or IP cover as AI models, may increase the risk of exposures;
- the potential for mitigation costs to respond to model retraining costs;
- the scope of cyber exclusions, for example whether ‘computer system’ is broad enough to encompass AI systems; and
- whether the ‘other insurance’ clause appropriately addresses how the policy will operate if the prospective insured also holds a bespoke AI policy that more appropriately responds to the same loss.
Considerations for bespoke policies
Where there is commercial appetite for bespoke AI risk policies or endorsements (known as ‘affirmative AI’ policies), the development of those wordings is something that will need to be kept under frequent monitoring as the AI landscape evolves. This includes from a technical product, use-case, liability and government policy perspective.
As jurisdictions take differing approaches to regulating AI’s development and adoption and assigning liability for harm resulting from AI (mis)use, bespoke wordings will need to be regularly reviewed to ensure they remain fit for purpose. London market wordings may also, over time, not be fit for purpose for insuring the AI risks present in other jurisdictions.
Underwriters should also bear in mind that insureds taking out bespoke AI cover will direct their AI liabilities there, rather than to financial lines policies that do not expressly cover AI risks. Therefore, where a bespoke AI policy or endorsement sits alongside products that respond to AI risks implicitly, the bespoke policy may, in practice, become the more collectable source of cover for insureds. Underwriters and claims handlers can then expect the bespoke AI cover to be considered first and should consider how ‘other insurance’ or non-contribution/excess clauses interact to determine priority of payment and contribution between the relevant Insurers. Bespoke policies will, therefore, need to be priced accordingly.
Considerations for claims handlers
While AI related liabilities have not yet become a common litigation theme, they are coming.
We expect that once initial ‘test cases’ exploring these issues have been successfully litigated, there will be a marked uptick in such cases (and insurance claims). Claims handlers will need to keep up with the different types of AI tools available and relevant industry guidance to be able to assess the extent that allegations against insureds have merit and the cost/benefit of defending claims. Until these liabilities have become an embedded part of caselaw, defence costs are likely to be higher than equivalent claims as Courts grapple with the applicable standard of care.
In practice, it would be worth considering the following when assessing notifications involving AI-related professional negligence:
- request details of the AI model that has been used;
- engage with underwriters about the disclosure of AI’s use by the business at the underwriting stage;
- consider whether the negligence in issue is human driven or AI model driven;
- consider the availability of bespoke AI policies and assess whether the notified policy implicitly provides cover for AI risks based on the current state of the law; and
- monitor emerging AI negligence ‘test’ cases and regulatory and professional bodies’ guidance closely.
AI claims are likely to raise a number of coverage issues that, whilst familiar, will need to be grappled with, including:
- boundaries of cover issues if wording is not tightly defined. For example, there may be professional liability cover for human error in not detecting the use of AI, but a problem with systems or software may more properly fall within a cyber policy. Similarly, if, for example, failure to maintain an AI system (or the security controls surrounding it) results in an AI system producing an erroneous result which leads to third party losses, there is potential for this to be regarded as a systems and controls issue that might fall to be dealt with under a management liability policy;
- notification considerations: when does an AI use become notifiable, for example, on early model failure or where there is a claim or loss. AI risks may evolve slowly, complicating timing and there may be questions over whether continuous AI errors constitute one or more circumstances;
- aggregation disputes compounded by difficulties in determining the cause of the act/omission at issue.
AI related professional negligence claims, and AI claims in general, have arrived and will continue to keep coming. The Insurers ahead of the underwriting, wording and claims handling implications will be best placed to manage the exposure going forward.
If you would like to discuss these developments and the implications arising, please do not hesitate to contact the team at the top of the page.
[1] The UKJT Paper expresses the view that establishing liability of upstream developers of AI tools (such as the model developers) will be difficult unless the resulting harm to the end customer was foreseeable.