这些 AI 模型的回答普遍存在计算失误、遗漏最新税务变动以及凭空捏造法规的情况，严重时可能导致用户面临重大的财务损失。例如免费版 Claude Haiku 4.5 在退休金税规上的幻觉回答可能导致储蓄者承担高达 17,500 英镑的税务罚款，另有模型误导留学生出国即可停止偿还学贷。尽管付费版本与具备推理功能的新模型表现相对较好，但最佳的 Claude Opus 5 推理模式仍有 39% 的出错率。
目前英国有约五分之一的成年人愿意让 AI 代为做出财务决策，年轻投资者对 AI 的信任度甚至高于传统理财网红或电视节目。然而，专业理财顾问警告，虽然聊天机器人可用于日常预算规划与初步资讯查询，但由于 AI 财务建议尚未纳入监管，消费者若因此遭受损失将无法获得任何赔偿保障，监管机构需加快步伐以保护大众利益。
Recent research conducted by technology firm Saturn reveals that relying on popular AI chatbots for financial guidance poses serious financial risks, with models frequently delivering inaccurate responses. Evaluating over 10,000 queries across ChatGPT, Claude, Copilot, Grok, and Gemini, the study found an average error rate of 57% on financial questions, which surged to 88%—and up to 99% in certain models—when handling complex multi-step calculations.
The models' responses frequently exhibited arithmetic errors, overlooked upcoming tax amendments, and hallucinated regulatory rules, potentially exposing users to severe monetary harm. In one documented instance, an erroneous answer by Claude Haiku 4.5 regarding pension tax rules could have resulted in a £17,500 penalty from HMRC, while another model falsely claimed overseas relocation exempts graduates from student loan repayments. Although paid and newer reasoning models demonstrated better accuracy, the top-performing Claude Opus 5 still produced errors in 39% of cases.
Despite these inaccuracies, approximately one in five UK adults remains open to allowing AI to make complex financial choices, with younger generations trusting AI more than financial influencers or broadcast media. Financial advisers caution that while chatbots can assist with preliminary research and personal budgeting, AI-generated financial advice remains entirely unregulated and lacks consumer compensation protections, underscoring the urgent need for regulatory action by financial authorities.