UK consumers warned over AI chatbots giving inaccurate financial advice (www.theguardian.com)

UK consumer group Which? tested 40 questions across major conversational AIs and found frequent, sometimes risky, inaccuracies in financial, tax and travel advice. Popular models gave misleading guidance: Microsoft Copilot and ChatGPT failed to flag a mistaken ISA allowance and could have encouraged oversubscription; ChatGPT falsely said travel insurance was mandatory for most EU countries; Meta returned incorrect flight compensation instructions; and Google’s Gemini suggested withholding payment from a builder, risking a breach-of-contract dispute. Which? scored Meta lowest, ChatGPT next, Copilot and Gemini slightly higher, and Perplexity—known for search—highest. Several models also surfaced links to premium tax-refund services instead of free government routes. The findings matter because increasing numbers of UK consumers (estimates from one in six to half) consult LLMs for money and legal matters, yet outputs remain prone to hallucination, stale data and prompt-sensitivity—evidenced when a deliberate incorrect ISA figure wasn’t caught. Regulators warn AI-generated advice isn’t covered by the Financial Ombudsman Service or Financial Services Compensation Scheme, exposing users to financial harm. Vendors say they’re improving safeguards (Google/Meta/MS/OpenAI point to user warnings and model upgrades such as GPT-5.1), but the tests underline a gap between consumer trust and current model reliability, and strengthen calls for clearer labelling, verification tools and possible regulatory action.