The Insurance Act 2015 (Insurance Act) imports the concept of the 'prudent underwriter'; a theoretical benchmark against which the adequacy of disclosure is assessed. A material circumstance is one that a prudent underwriter would consider relevant to the risk. An insurer is taken to know matters that a prudent underwriter would know. 

As AI-powered underwriting tools become industry-standard, the question arises: Does the prudent underwriter standard evolve to incorporate AI-assisted analysis?

If a market-standard underwriting platform would routinely mine publicly available data such as social media posts, Companies House filings, satellite imagery and news feeds to identify risk factors, can an insurer later claim ignorance of matters that such a platform would have uncovered? There is a credible argument that the answer may in some circumstances be ‘no’.

Insurers could be deemed to know what an AI system could reasonably have discovered, even in the absence of disclosure by the insured. This would represent a significant doctrinal shift, and one that the courts have not yet had occasion to specifically address.

Reasonably clear and accessible requirement 

The Insurance Act requires disclosure in a particular manner. It provides that a fair presentation is one (amongst other provisions of the Insurance Act) that: 

“Which makes that disclosure in a manner which would be reasonably clear and accessible to a prudent insurer” (Section 3(3)(b) of the Insurance Act).

“The disclosure required is as follows, except as provided in subsection (5):

(a) disclosure of every material circumstance which the insured knows or ought to know, or

(b) failing that, disclosure which gives the insurer sufficient information to put a prudent insurer on notice that it needs to make further enquiries for the purpose of revealing those material circumstances.” (Section 3(4) of the Insurance Act).

“(5) In the absence of enquiry, subsection (4) does not require the insured to disclose a circumstance if:

(a) it diminishes the risk,

(b) the insurer knows it,

(c) the insurer ought to know it,

(d) the insurer is presumed to know it, or

(e) it is something as to which the insurer waives information." (Section 3(5) of the Insurance Act).

This was introduced to prevent insureds from complying with disclosure obligations by providing insurers with a mass of documents and leaving it to the underwriter to identify what is material. This was designed to stop insureds overwhelming insurers with information in a way that obscured material information, often known as ‘data dumping’. 

The use of AI in underwriting creates an interesting complication. The Insurance Act was drafted before the widespread use of AI in underwriting. Where underwriting decisions are made in whole or in part, by AI, the ability to identify material disclosures buried within a lengthy submission is critical. There is an argument that the "reasonably clear and accessible" standard may shift with the use of AI in the industry.

The courts might accept that AI-assisted underwriting processes set a different benchmark for what is "accessible to a prudent insurer" now that many underwriting decisions are made in whole or in part by AI, as to identify material disclosures buried within a lengthy submission has shifted. This could be problematic if a particular insurer’s use of AI is not keeping up with the market or if their AI fails to identify material information buried in a document dump. It may impact an insurer’s ability to rely on non-disclosure and assist arguments that it failed to make reasonable inquiry.  

Intelligence gathering during the policy period

Insurers are increasingly deploying AI to monitor risks on a continuous basis throughout the policy period. This may include scanning news sources, tracking claims trends, and monitoring policyholders' public-facing activities. Where such monitoring reveals a material change in risk, questions arise as to whether the insurer has an obligation to act on that intelligence, and how mid-term disclosure obligations interact with this new capacity for real-time awareness.

What does this mean for insurers? 

Where an AI-powered proposal platform fails to capture, process, or flag a material piece of information correctly, the consequences for the validity of the policy could be significant. AI could change the constructive knowledge that insurers are considered to have. Insurers should be alive to the risk that technical failures in AI systems or failure to keep up with the market could inadvertently generate disclosure issues. This has potential consequences for cover and claims handling.

The existing legal framework, designed for a world of human- only underwriting may not remain adequate with the increasing use of AI. A more systematic legislative response may be needed that provides clear guidance on the scope of constructive knowledge where AI systems are deployed. 

As AI reshapes what it means to 'know' a risk, and what it means to 'present' that risk fairly, the insurance market must engage seriously with the questions it raises.