Financial advice from AI chatbots like ChatGPT and Claude frequently proves incorrect, a study finds, risking thousands in unexpected tax bills.
AI chatbots fail financial advice tests
Tech research firm Saturn conducted extensive tests on 18 prominent AI models, including ChatGPT and Claude. They posed 121 distinct money-related questions to each model five times, accumulating over 10,000 responses for analysis.
The study’s findings revealed significant inaccuracies. Chatbots provided answers that were either incorrect or materially incomplete in 57% of all cases. This suggests a widespread reliability issue across various AI platforms.
For more difficult financial questions, well-known tools like ChatGPT, Claude, Grok, and Gemini performed even worse. They made mistakes in 88% of these complex scenarios, with some models failing 99% of tests. Free models consistently performed worse than paid alternatives.
Chatbot financial advice error rates
A study by tech research firm Saturn highlighted substantial error rates in financial advice provided by AI models. These figures indicate a significant risk for users seeking monetary guidance.
- Overall inaccuracy: 57% of all responses were either inaccurate or materially incomplete.
- Complex questions: Error rates soared to 88% for more difficult financial queries.
- Free vs. paid models: Free AI tools gave incorrect answers in 63% of cases, while paid alternatives had a 49% error rate.
- Worst performer: Claude Haiku 4.5 made mistakes in 82% of its answers, the highest error rate among tested models.
- Second worst: Google’s Gemini 3.1 Pro failed 73% of tests, indicating similar reliability issues.
- Specific risk: An incorrect answer from Claude regarding pension tax rules could have resulted in a GBP17,500 tax bill from HMRC.
FCA warned on AI financial advice risks
The Financial Conduct Authority (FCA) had already voiced concerns in August, publishing research on public understanding of AI financial advice. This earlier study identified a significant gap in consumer awareness regarding the risks involved.
The FCA’s findings showed that 44% of those surveyed mistakenly believed general-purpose AI chatbot information was regulated. Such advice falls outside FCA oversight and consumer protection.
Furthermore, 38% of respondents considered it acceptable to base investment decisions solely on AI-generated data. Many also wrongly thought they could seek compensation for losses incurred from AI advice.
Users face significant financial losses
The most immediate consequence is the potential for substantial financial loss. An incorrect AI response on pension tax rules, for instance, could lead to an unexpected GBP17,500 tax bill from HMRC.
Saturn’s chief executive, Amal Jolly, warned that millions are trusting AI for money advice, often receiving incorrect information that costs them money. This widespread reliance on unreliable tools creates a broad risk.
The FCA spokesperson reiterated that AI is not a “crystal ball” and does not provide regulated financial advice. Consumers must always verify information and prioritise long-term financial goals over quick-rich opportunities.
How will AI models be regulated?
The most pressing question is how regulatory bodies, like the FCA, will adapt their frameworks to address the rapidly evolving capabilities of AI models offering financial information, which currently operate largely unchecked.
- Will general-purpose AI models be brought under existing financial regulation, or will new, specific guidelines for AI-generated financial advice be developed?
- How will AI developers be held accountable and compelled to improve the accuracy of their financial responses, particularly for complex queries?
- What clear recourse or compensation mechanisms will be established for consumers who suffer financial losses due to AI-generated misinformation?
Wider concerns over AI reliability
The issues with AI financial advice are not isolated; they mirror broader concerns about AI models struggling with factual accuracy across various domains. This pattern suggests a fundamental limitation in current general-purpose AI.
Instances of AI generating fabricated legal case law have been reported, and medical AI tools face scrutiny for potential diagnostic errors. These examples highlight the pervasive challenge of AI reliability.
The core problem lies in AI’s capacity to produce highly convincing but entirely false information. This makes it exceptionally difficult for users to discern truth from fabrication, especially in technical fields like finance or law.
Financial AI needs urgent regulation
Saturn’s research delivers a clear message: general-purpose AI models are currently unsuitable for providing financial advice. The risk of consumers incurring substantial losses, such as a GBP17,500 tax bill, is simply too high to ignore.
The rapid advancement of AI has undeniably outpaced existing regulatory structures. The widespread misconception among consumers that AI financial advice is regulated exposes a critical void in both oversight and public education.
Regulators must establish clear and enforceable guidelines for AI models that offer financial information. This could involve specific licensing, mandatory disclaimers, or even restrictions on the types of advice unregulated AI can provide.
While regulatory action is paramount, individual users also have a responsibility. They must approach AI-generated financial guidance with extreme scepticism, understanding its inherent limitations and the complete absence of consumer protections.
Verify all AI financial information
Anyone considering using AI chatbots for financial planning must understand that such information is not regulated. This means there are no consumer protections if the advice leads to financial losses.
You cannot expect compensation from schemes like the Financial Services Compensation Scheme or redress through the Financial Ombudsman Service for losses incurred from AI advice. Always cross-reference AI advice with trusted, regulated financial professionals or official sources.
The FCA strongly advises consumers to verify all information and remember that past performance does not predict future returns. Be highly sceptical of any AI suggestions promising quick wealth, as these are often unreliable.
