AI agents are increasingly becoming part of tax due diligence workflows, helping professionals analyse transaction data, identify potential exposures and summarise complex regulatory information. But as these systems become more embedded in decision-making, a new risk is emerging: AI that agrees too readily.
TAINA Technology‘s recent analysis by Rich Kent delved into the risks of sycophantic AI and why organisations adopting AI agents in tax due diligence need to prioritise challenge, governance and professional scepticism.
The issue is not whether AI can produce answers. It is whether those answers challenge assumptions, test evidence and identify potential weaknesses. In tax due diligence, where the purpose of analysis is to uncover risk rather than confirm expectations, an AI system that reinforces user assumptions can create a false sense of confidence.
Sycophancy describes a tendency for AI systems to prioritise agreement over accuracy. Instead of independently evaluating information, an AI agent may mirror the assumptions contained within a user’s question, producing responses that validate an existing view rather than testing whether it is correct.
For tax professionals assessing a potential transaction, this creates a significant challenge. If an AI agent is asked whether a particular tax position appears low risk, the way the question is framed can influence the response. A system designed to be overly agreeable may reinforce the user’s initial conclusion rather than investigate alternative interpretations or highlight areas requiring further review.
This matters because the value of tax due diligence comes from identifying what has been missed. The role of the process is not to confirm that a transaction appears sound, but to uncover exposures that could affect valuation, compliance or future liabilities.
An AI agent that consistently validates conclusions can amplify confirmation bias, reduce professional scepticism and make potential risks easier to overlook. In highly regulated environments, confidence without sufficient challenge can be as damaging as a lack of information.
However, sycophantic behaviour can often be identified through testing. Organisations should monitor whether an AI agent regularly agrees with initial assumptions, struggles to provide alternative perspectives or changes conclusions when the same scenario is presented differently.
A useful test is asking an AI system what evidence would change its conclusion. If it cannot identify opposing factors, limitations or circumstances where its assessment may be incorrect, additional scrutiny is required.
Managing this risk requires a different approach to AI deployment. Rather than positioning AI agents as validators, organisations should use them as challengers, asking questions such as what assumptions are being made, what risks may have been overlooked and what alternative interpretations should be considered.
This does not remove the need for human expertise. In tax due diligence, professional judgement, oversight and accountability remain essential. AI should support professionals by challenging thinking and surfacing potential issues, not replace the critical assessment required to make informed decisions.
TAINA’s analysis highlights a broader lesson for organisations adopting AI agents across compliance workflows, that the most valuable systems will not be those that simply provide answers, but those capable of questioning them. As AI becomes more embedded in professional services, the ability to challenge assumptions may become one of its most important functions.
Read the full Taina Technology post here.
Copyright © 2026 FinTech Global
