Revolut is using OpenAI models and Codex in software development, security investigations and the detection of financial crime. The company says AI allows it to screen more transactions while giving engineers more time to design systems and review solutions.
Revolut operates in more than 40 markets and serves 800,000 businesses; its consumer customer base has grown by 23% over the past 12 months. Kim Maniovich, Revolut’s Global Head of Business Marketing, says the partnership with OpenAI helps the company develop solutions with small teams, manage costs and adapt tools to its own ways of working.
Financial crime and growing demand for investigations
As the company grows, protecting customers becomes more complex. A report cited in the article found that more than half of adults worldwide encountered a scam in 2025. At Revolut, transaction volumes far exceed what human investigators could review individually.
David Abitbol, Revolut’s AI Lead, says one of the biggest challenges is closing the gap between human capacity and rising demand. The company therefore uses a FinCrime agent powered by OpenAI models. It examines information, applies Revolut’s rules and automatically completes the defined investigation process for eligible cases.
Abitbol says the release of GPT-5.5 marked a turning point: the models had become capable and cost-effective enough for Revolut to delegate screening and investigation workflows entirely to the agent. This allows the company to screen more transactions without increasing investigator headcount at the same rate as demand. Information Security Engineer Mo Amiri says the approach provides stronger customer protection and a more resilient system.
Faster development with human accountability
According to Abitbol, Codex has reduced coding time by 30–60%, measured from the start of a task to the merging of a feature. The company applies common standards to the development, A/B testing and prediction of business metrics for systems built with AI.
Senior Platform Engineer Dmitry Pitik experienced the change while building an internal code-review service. After designing the system and breaking the work into features, he delegated implementation tasks to multiple Codex agents rather than requiring a full development team. He concentrated on design, planning and reviewing the results.
Revolut says the time saved gives engineers more opportunity to discuss design decisions. Pitik stresses, however, that developers remain responsible for the code they produce and do not trust AI blindly.
The company is using the resulting capacity elsewhere, too. It has developed a tool for building web interfaces for internal applications and external-facing products. It also converted roughly seven years of design-system documentation into AI-readable context in a single day. Senior Designer Dmytro Pavlov says this can provide a foundation for tools that help designers work and make accumulated organisational knowledge more accessible.
Security investigations and review
Amiri’s work includes ensuring that faster development does not come at the expense of security. In a phishing investigation, an analyst previously had to collect the email, inspect links and attachments, trace the sender and assemble the evidence. AI can help uncover connections and present a structured case, but the investigator must still check the original records, assess the explanation and approve any action.
When reviewing new AI applications for security, Amiri also examines what information a system can see, where data goes, which tools it can access, what it can do with them and what happens if it fails. He checks whether apparently convincing conclusions are supported by evidence. Codex saved one to two days on one substantial security code-review task.
In the company’s approach, speed does not replace accountability: if an AI tool suggests a code change, an engineer must understand it well enough to stand behind it.
Potential customer-facing applications
Alongside internal uses, Revolut sees possible applications for AI in everyday financial tasks. Abitbol outlined the possibility of a conversational interface through which customers could, for example, block a card, manage subscriptions or look for ways to save for a trip.
Pavlov says the associated design challenge is to turn complex processes into an interface that is consistent, familiar and easy to use. Maniovich hopes to see the company’s ambitious working culture reflected in the products it builds for customers. These are future possibilities, however, rather than announced finished features.



