Anthropic has launched a cluster of AI plug-ins targeting financial services - investment banking, private equity, equity research and wealth management. The tools promise to review transaction documents, build competitor analyses, parse earnings transcripts and update financial models in real time. Some investors already use them for market research, deal sourcing and drafting early-stage investment memos. But the people closest to the money say the technology has a fundamental problem.

"It's absolutely crap at judgment," Jean-Baptiste Brian, co-chief executive of private equity firm Hg Capital, told the Financial Times. Brian described AI tools as "sycophantic" - agreeing too readily with whatever hypothesis the user presents. "It can't do investing right now," he said, "Hvylya" reports.

The concern extends beyond private equity. Sam Jones, chief operating officer of Compare the Market, said plugging AI into an insurer is technically possible but "generates pretty catastrophic consumer outcomes." Jones warned the technology "can hallucinate on consent, it can hallucinate on terms and conditions, it can hallucinate on prices" - failures that could expose companies to regulatory action and erode customer trust.

Jones said Compare the Market is building a "secure harness" around AI models to ensure outputs meet the accuracy and transparency standards that customers and regulators expect. The approach reflects a broader pattern across financial services: companies eager to adopt AI capabilities but unwilling to hand over final decision-making to a system that cannot be held accountable for errors.

Victor Englesson, a partner at investment firm EQT Partners, said top-line growth has been slowing in parts of the software sector for some time, and AI has amplified the pressure. To prove public markets wrong, he argued, software groups must show customers are willing to pay for new AI features - not just use them as free add-ons. The technology is powerful, Englesson said, but "who is best positioned to capture that value" remains the open question.

Also read: why FT columnist Martin Wolf warned that AI poses a deeper threat to the educated class than previous waves of automation.