Jeder zweite Finanz- und Versicherungsbetrieb in Deutschland nutzt generative KI.

Finance already runs on AI

In two years, the share of Japanese financial institutions using generative AI rose from 31% to 79%, the Bank of Japan reports from a survey of 150 institutions. Three in four UK financial firms used AI in 2024, and foundation models already made up 17% of their use cases (Bank of England and FCA). Nearly two in three European insurers actively use generative AI, most still at proof-of-concept stage, across 957 reported use cases (EIOPA, 2 Feb 2026). In Germany, half of all finance and insurance establishments used generative AI in 2025, twice the average across the economy (IAB Kurzbericht 8/2026).

That is the starting point of the Good AI Prism Sector Brief: Financial Services & Insurance. The question is no longer whether banks and insurers use AI. It is whether they can prove it is accurate, fair and controlled, to supervisors, auditors and customers.

Fraud and process first, service next

Internal process optimisation (41%), cybersecurity (37%) and fraud detection (33%) lead today among UK firms, and firms expect the largest growth in customer support and compliance. The stakes are high: fraud scams and bank fraud cost an estimated $486bn worldwide in 2023 (Nasdaq Global Financial Crime Report).

Work changes before headcount does

Finance employs large, experienced workforces facing a wave of retirements; Germany's insurers alone employ 215,600 people (AGV Versicherungen). The European Central Bank finds no significant difference in hiring or firing between firms that use AI and those that do not; firms that use AI intensively are slightly more likely to hire (ECB Blog). Customers move faster than many institutions expect: a quarter of Germans have already asked an AI tool about money, a third of those aged 16 to 29, and 62% say AI raises their fear of fraud (Bitkom, Digital Finance 2026).

The rules: high-risk duties for credit scoring and pricing from Dec 2027

From 2 Dec 2027, creditworthiness assessment and credit scoring of natural persons, and risk assessment and pricing in life and health insurance, carry high-risk duties under the AI Act, with dates as amended by Regulation (EU) 2026/1744. Fraud detection is explicitly excluded. Deployers run a fundamental-rights impact assessment. Before that, DORA has applied since 17 Jan 2025 to ICT risk, incidents and third parties, and from 20 Nov 2026 the Consumer Credit Directive 2 lets consumers ask for human intervention and an explanation when credit is assessed automatically.

Supervisors elsewhere set expectations by guidance. In the US, the revised interagency model risk guidance replaced SR 11-7 in April 2026 and leaves generative and agentic AI outside its scope. 29 US jurisdictions have adopted the NAIC model bulletin on AI in insurance. Canada's OSFI Guideline E-23 covers AI models from 1 May 2027, and Singapore's MAS guidelines add board oversight, an AI inventory and lifecycle controls. Australia's regulator found governance lagging AI use at some licensees (ASIC REP 798).

What the market offers

The Good AI Prism places 28 offerings in Financial Services & Insurance. Eight fight fraud, financial crime and money laundering, among them Feedzai, Hawk AI and BioCatch. Five support credit decisions, such as Zest AI, Taktile and Provenir. Five handle claims and claims fraud, including Shift Technology and FRISS. Four price and underwrite insurance risk, such as Akur8 and Earnix. Three serve research, among them AlphaSense, and three power banking assistants and personalisation. 15 of the 28 offerings come from providers headquartered in Europe, three of them in Germany. The Financial Services & Insurance area shows every entry with its source and date.

What to do Monday

1. Start an AI register: every model and AI use, including generative tools staff bring themselves, with purpose, owner and risk class.
2. Mark the uses that score people or price life and health cover, and build explanation, human review and fairness testing in before 2 Dec 2027.
3. Extend model risk management to language models and agents, with validation, monitoring and change control.
4. Check that every AI provider sits in your DORA register of ICT third parties, with exit and incident plans.
5. Pilot one fraud, claims or service use with a baseline and fairness tests, and scale only on measured losses, accuracy, handling time and customer outcomes.