Consider a scenario familiar to many who work in or alongside a document-heavy function. A compliance report is circulated for sign-off, drafted from last quarter's template with a new paragraph and different dates.

The new paragraph gets read, the dates get confirmed, and everything is approved. A document from a junior colleague gets skimmed, an item or two gets flagged and out it goes.

These examples reflect a real challenge in today’s commercial world: high-stakes pressure and finite capacity.

Many professionals in Bermuda's corporate sector are asked to review more material, faster, with the same or fewer people. Review becomes a procedural step rather than a careful understanding of the issues and risks.

Now, artificial intelligence tools are generating first drafts of the very documents that professionals are responsible for developing and reviewing. The volume of draft material has increased sharply but review capacity has not.

In many professional services teams, junior colleagues produce first drafts that seniors review, so what difference does it make if the draft came from a person or a tool? The difference is psychology.

A junior professional's draft often looks like a draft and invites scrutiny. AI-generated documents, by contrast, look polished, well-structured, and confident, and senior professionals are more ready to accept an AI’s draft following a cursory browse.

The reviewer's instinct shifts from “what is wrong with this?” to “this looks right”.

The Bermuda Monetary Authority’s consultation paper on the responsible use of AI in Bermuda's financial services sector identifies this exact dynamic. The paper names “automation bias” as a distinct AI-related risk, noting that “GenAI can produce plausible but inaccurate outputs”.

Often, these AI outputs show no obvious errors, but well-formatted, believable inaccuracies that a reviewer skimming under time pressure will be less likely to catch. If an AI output misreads a regulation, the polish of the output can make it feel like the AI “author” knew what it was talking about.

There’s also a growing practice of feeding dense documents into AI and relying on its summary. It takes time to review and understand large documents, and time is a scarce resource. Depending on the complexity and importance of the document, the difference between proper review and AI-assisted recaps could be the difference between catching a buried obligation and missing it entirely.

The consequences of this approach are material. When a compliance report or a regulatory filing is produced with AI assistance and reviewed superficially, or important documents are summarised by AI without independent scrutiny, decisions may be based on analysis that is wrong or misconstrued.

Then there is the audit trail. When work is generated through AI tools, modified informally and finalised without documented review steps, the audit trail is much harder to follow.

If a regulator or auditor asks how a particular output was produced and who checked it, the answer may be that no one is sure and which company wants to explain that to a regulator?

A good starting point to manage this issue is diagnosis. Pick a sample of recent outputs from the legal, compliance, or governance function and ask: what was checked, how long did the review take, and is there a record? For outputs produced with AI assistance, include verification of factual and regulatory claims against source authorities, identification of any content that may be hallucinated or inaccurate, and a clear record that these steps were taken.

The BMA's consultation paper gives Bermuda's regulated entities a proportionate, practical framework.

But the framework’s effectiveness also depends on the people reviewing the outputs applying scrutiny, not just the process of review itself.

If your team's most recent regulatory filing or compliance report was challenged tomorrow, on accuracy, analysis, and whether anyone scrutinised it before it went out the door, could you defend the review that was done? If not, the time to fix that is now, before someone comes asking.

• Christa Schweizer is the principal consultant, legal and governance, at Risk Intelligenz, a British-based consultancy