AI agents are gaining attention across finance, but banks still lack the trust systems needed to hand them real financial authority.
Interest remains high, yet adoption is moving slowly because institutions must protect money, compliance, client data, and accountability. The debate has shifted from model power to control, memory, and auditability.
Citi Sky Highlights Banking Limits
Citi’s launch of Citi Sky on April 22 gave the market a clear example of the challenge. The AI wealth assistant was built with Google Cloud and Google DeepMind technologies and uses Google’s Gemini Enterprise Agent Platform. Citi plans a phased rollout to Citigold clients in the United States this summer.
Dipendra Malhotra, Citi’s wealth technology head, pointed to memory as a core weakness for high-stakes advisory tools. He questioned how long a client conversation can continue before the system begins to lose context or produce unreliable answers.
That concern matters in wealth advice, treasury management, and portfolio execution. Retrieval systems can extend memory through outside databases, but context limits still shape how much an agent can handle at once. In banking, that limit becomes an operational risk.
Ethereum Standards Target Agent Trust
Ethereum developers are drafting standards that could help autonomous agents prove identity, build reputation, and complete tasks under clearer rules. ERC-8004 proposes three registries covering identity, reputation, and validation. These systems aim to let agents show who they are and how they have behaved.
ERC-8183 focuses on escrowed jobs and evaluator attestation. Under that model, a client funds a task, a provider submits work, and an evaluator accepts or rejects the result. The proposal does not settle disputes, but it offers a framework for verifiable completion.
Still, reputation remains difficult. Agents can create activity faster than humans, which may inflate trust signals. Banks must decide whether an agent’s record shows real reliability or only repeated automated actions.
Finance Still Faces Accountability Gaps
A Deloitte poll of more than 3,300 finance and accounting professionals showed strong interest but low adoption. About 80.5% said AI tools such as agents and GenAI chatbots could become standard within five years. Only 13.5% said their organizations already used agentic AI.
McKinsey estimates that 50% to 60% of bank full-time equivalents are tied to operations. That makes finance a major target for automation, yet many institutions remain stuck in limited pilots instead of changing operating models.
Industry leaders say financial agents must stay within user instructions, allow users to stop them, and avoid moving assets to third parties. The unresolved issues remain responsible for losses, trustworthy reputation, operational control, and regulation when agents act outside their scope.
