AI won’t void your firm’s PII, but it will shape your renewal

By Kenny MacAulay, CEO of Acting Office

At the beginning of October, the ICAEW updated its PII regulations, including a minimum approved wording policy which covers any civil liability incurred in professional work. Under these new guidelines, cover attaches to the liability and not the AI tool, which means AI errors are covered in the same way as any other professional error.

Any policy update should aim to clarify guidance, making regulation easier to implement for firms. That being said, the wording is silent on AI, which leaves firms unsure what they need to tell their insurer about it.

In practice, the wording shifts the argument between firm and insurer to proposal and renewal. That’s where firms will be judged on what they disclose about their AI use, how underwriters price it, and whether they can prove their controls work.

Insurers are already alert to this. Out of 144 Lloyd’s managing agents surveyed by the Lloyd’s Market Association about which insurance lines are most likely to be impacted by AI, professional indemnity came out on top. Managing agents are the firms that underwrite business at Lloyd’s, so this is the view of the people pricing firms’ cover. This raises the question of how much trust accountants can put in AI tools, and crucially how they evidence the checks made before false content reaches a client.

Where the claim starts

In theory, if an AI-assisted output is included in a client report, this action inherently creates client reliance on AI-generated information. However, if this output is later proven to be false, the liability subsequently falls onto the accountant. A firm might have some comeback against its AI provider under contract, but that won’t stop the client suing the firm. The engagement was with the firm, and so was the advice.

Deloitte Australia found this out last year. It refunded more than A$97,000 of a A$440,000 fee to the Australian government after a report it delivered was found to contain a fabricated quote from a federal court judgment and references to academic papers that did not exist. A revised version disclosed that a generative AI model had been used in its preparation, although the firm did not attribute the errors to it. The client did not pursue the software vendor. It went to the firm whose name was on the report.

That, however, is only one report from one engagement. The truth is that not every business is aware of its total internal AI usage across the whole firm. Therefore, claims become more difficult to field as AI usage is not always monitored or detectable when a claim occurs.

Recently, BlackFog found that 47% of security leaders report sensitive data exposed via an AI tool, while 77% of those involved an AI tool that was pre-approved by the company.

What underwriters are already asking

Underwriters will need to know the exact models that are already in use within the company. From there, they will investigate the justifications behind their usage and how or even if these models are being surveyed or regulated.

This information should be gathered and policies implemented before a client becomes involved to avoid potentially damaging claims.

Firms that are unable to verify answers to these questions may face harder renewal and tighter terms.

The choices facing firms

There is, however, no single solution to avoid the dangers of potential AI-based PII claims. Firms face four choices, and each does a different job.

The first is an AI register, a record of every AI tool in use across the firm, who uses it and for what. This is usually the first thing an underwriter will ask for.

The second is human review, a documented check of AI-assisted work before it reaches a client. Its value at renewal depends on whether the firm can show the review took place.

The third is disclosure to clients. Firms can disclose their AI usage from the outset via engagement letters, being transparent about the level of their reliance on AI outputs. This includes naming the models, highlighting artificially generated information or research, and to what extent humans verify these outputs.

As a result, firms are able to set clear expectations from the start, although this does not remove liability. A client who has been told AI is used is still entitled to accurate work.

The fourth is disclosure to insurers. Full and accurate disclosure at proposal and renewal is what allows a firm to evidence its controls, and puts it on firmer ground if it ever needs to make a claim. Which combination is right will depend on the firm, but none of them can be left out of the renewal conversation.

spot_img
spot_img

Subscribe to our Newsletter