trending
Citi has introduced Arc, a new internal platform designed to help the bank build, govern and scale artificial intelligence agents across its global business.
Citi builds an operating layer for AI agents in banking
The launch marks a further step in the shift from generative AI as a productivity tool towards agentic systems capable of completing defined tasks inside regulated financial institutions.
The platform will allow Citi developers to create AI agents for specific use cases, with the bank initially focusing on tightly controlled applications before expanding access more widely.
Citi says the agents will support activities such as research, synthesis, preparation and execution, reducing manual work while leaving human judgement at the centre of decision-making.
From assistance to execution
The distinction matters. Many banks have already deployed AI tools to help staff draft documents, summarise information or search internal knowledge bases.
Arc appears to move beyond that model by giving Citi a structured environment in which multiple agents can be embedded into business workflows.
One example cited by the bank is wealth management. A banker preparing for a client meeting might currently spend hours gathering portfolio information, analysing market trends and modelling possible scenarios.
Citi envisages AI agents completing much of that preparation proactively, delivering relevant material when needed and allowing the banker to focus on advice, relationship management and client outcomes.
That reflects a broader change in the role of banking staff. Rather than acting primarily as coordinators of information, employees may increasingly become reviewers, advisers and architects of decisions supported by machine-generated analysis.
Governance will determine credibility
For large banks, the challenge is not simply whether AI agents can perform useful work. It is whether they can do so in a way that is monitored, auditable and compliant with regulatory expectations.
Citi has emphasised that agents developed through Arc will be governed and observable, giving the bank visibility over what they are doing, how they are operating and what value they are creating.
That governance layer is essential. Agentic AI inside banking raises questions around accountability, data access, model risk, customer treatment and operational resilience.
A poorly controlled agent could create errors at speed. A well-designed one could remove friction from repetitive processes while improving consistency and responsiveness.
Citi’s existing AI adoption gives the project a sizeable foundation. The bank says more than 80 per cent of employees with access to its in-house AI tools use them regularly, while most have completed prompt training.
A wider race to agentic finance
Arc also sits within Citi’s broader investment in AI infrastructure, data modernisation and client-facing experimentation, including recent work with Google on an AI-powered avatar for wealth clients.
The timing is significant. Banks, payment companies and retailers are all exploring how autonomous or semi-autonomous AI can reshape commerce, financial advice, fraud prevention and back-office operations. Yet adoption will depend less on novelty than on trust, transparency and control.
Citi’s bet is that agentic banking will not be delivered through isolated AI experiments, but through a governed platform model. If Arc succeeds, it could become part of the architecture through which large banks turn AI from a workplace assistant into a regulated operating capability
Comments