The UK Financial Conduct Authority has warned that AI oversight in finance is turning into an “arms race” and said lawmakers should give it new powers to supervise large models including ChatGPT, Claude and Gemini. In its July 6 warning, the FCA argued that generative AI is moving quickly into banks, insurers and asset managers while supervisory tools for detection and validation are not keeping pace.
Regulator says financial decisions are becoming harder to inspect
According to the Financial Times, the FCA used the “arms race” analogy to describe the widening gap between industry adoption and regulatory capability. Each added layer of AI inside a financial institution requires a matching layer of oversight tools if regulators are to test accuracy and review decision paths. The concern is simple. Traditional credit assessments in banks relied on statistical models, and insurance pricing was built on actuarial formulas, but those channels are now being replaced by large language models that sit outside the FCA’s current statutory reach.
Three major models named in the call for oversight
The regulator specifically referred to ChatGPT from OpenAI, Claude from Anthropic and Gemini from Google. The report said these systems differ in training methods and data sources, which is why the FCA wants a benchmark testing framework behind them. The idea is to check whether outputs used by financial firms meet regulatory standards before deployment. The issue is not brand recognition; it is whether model behavior can be examined in a consistent way.
Rising AI spending adds to the pressure on regulators
On the same day, The Kobeissi Letter said AI capital expenditure by Alphabet, Amazon, Meta, Microsoft and Oracle is projected to reach $1.1 trillion by 2027, exceeding US defense spending for the first time. Spending by those five companies as a share of US GDP is expected to rise from 1.5% in 2025 to 3.2% in 2027. The money going in is large, and the influence of AI in financial decisions is expanding with it, from asset allocation to risk management.
Taiwan comparison highlights a broader policy gap
The source also pointed to Taiwan, where Cathay, Mega, Shin Kong and Taishin have publicly announced AI customer service and robo-advisory deployments. Based on 2025 data cited in the report, IT spending growth in the banking sector reached 18%, with AI-related budgets accounting for more than 35%. Compared with the UK, Taiwan’s current framework was described as still relying on self-declaration, without mandatory model testing rules. If deployment keeps moving faster than oversight, the same “black box” decision risk identified by the FCA could surface elsewhere as well.