Sibos settled the technology debate. It has not yet faced the identity debate that comes next.
Miami solved half the problem
If you were at Sibos in Miami last month, you heard one word more than any other: trust. Bank CEOs used it about markets. A Federal Reserve governor used it about liability. Card networks used it about whether there is a real customer behind an agent.
The industry left Florida agreeing on something it has circled for three years. The technology for AI agents is no longer the argument. Banks are running agents in production today, with deterministic guardrails around them and people on the loop. The question now is who carries the risk when an agent acts, and whether it was ever entitled to act at all.
That is the right question. But most of the answers we heard were about the agent: better harnesses, tighter controls, clearer audit trails. All necessary. None sufficient. Because the weakest link in agentic banking is not the software making the decision. It is the bank's knowledge of who that software is working for.
Every mandate needs a principal
The vocabulary on many stages at Miami was intent, mandate, and accountability. What did the human intend when they set the agent loose? What limits does it operate within? Can the bank prove, after the fact, that each action matched the instruction?
In retail, the principal is a person with a phone and a biometric. In corporate and institutional banking, the principal is a legal entity. It has a registered address, a board, beneficial owners several layers up a group structure, authorized signatories, and a sanctions profile. Any of those can change on a Tuesday afternoon.
So, when a treasury agent instructs a payment, rebalances liquidity or opens an account for a new subsidiary, the bank must answer three questions in the time it takes to process the request. Is this entity who it says it is? Is the person who delegated authority still entitled to delegate it? Has anything changed in the ownership or risk profile since we last looked?
A mandate written against stale identity data is not a control. It is a liability with a timestamp.
KYC was built for a slower economy
Here is the uncomfortable truth. Most banks still know their corporate clients through a file. It was assembled at onboarding, stitched together from PDFs, registry extracts, and emails, and refreshed on a one, three or five-year cycle depending on risk rating. Between reviews, the bank is operating on memory.
That model was tolerable when a human relationship manager sat between the client and every material decision. It breaks completely when software acts on the client's behalf thousands of times a day. An agent cannot phone a colleague to check whether a director resigned last month. It acts on what it is given.
One of the sharpest lines in Miami came from HSBC: "You don't AI a bad process." Fix the process and the data first, then put AI on top. Apply that to agentic corporate banking and the conclusion is unavoidable. You cannot hand an agent a stale KYC file and call it governance.
The industry has spent a decade treating KYC as a compliance cost to be minimized. In an agentic economy, it becomes the foundation every automated decision stands on.
Identity is the agent's license to operate
The banks that win the next phase will stop thinking about client identity as a document and start treating it as infrastructure. That means a corporate digital identity that is structured, machine-readable, sourced from authoritative registries and continuously monitored. Not a snapshot. A live feed.
With that in place, the mandate questions become answerable at machine speed. The agent knows which entity it represents. The bank knows who delegated authority and whether that authority still stands. A change in beneficial ownership or a new sanctions hit adjusts what the agent is permitted to do before the next instruction lands, not at the next periodic review.
This is also where the liability debate gets resolved. Regulators will not ask whether your agent was clever. They will ask whether you knew who your client was at the moment the agent acted, and whether you can prove it. A bank with live, evidenced identity data can answer yes. A bank with a three-year-old file cannot.
There is a commercial point too. Mastercard argued in Miami that banks will not get to choose whether they take part in agentic commerce, because clients will hand credentials to agents regardless. If that holds for corporates, the bank that can verify the entity behind an agent instantly becomes the counterparty of choice. The bank that cannot becomes the bottleneck client’s route around.
Three questions for the board
If you sit on an executive committee approving an agentic AI roadmap, ask these before you sign off the next pilot.
Who owns client identity as a product? In most banks it is split across onboarding, compliance, operations, and the front office, and owned by none of them. Agents need a single, authoritative source. Someone senior must be accountable for it.
How old is the data your agents will act on? If the honest answer is "as old as the last periodic review," your agentic strategy is resting on sand. Move to event-driven refresh, where a change at the registry triggers a change in the bank.
Can you evidence every decision back to source? Agent audit trails are only half the story. You also need provenance for the identity data the agent relied on: which registry, which document, which date. That is what will stand up in front of a supervisor.
Speed belongs to the banks that know their clients
I am bullish on agentic banking. The productivity gains are real, and the client demand is real. The banks describing agents in production in Miami are not experimenting anymore; they are compounding an advantage.
But the next leap will not come from a better model. It will come from banks that can tell an agent, with certainty and in milliseconds, exactly who it is acting for and what that entity is allowed to do. Trust at agentic speed is not a governance slogan. It is a data architecture decision, and boards should be making it now.
Sibos moves to Singapore next year. By then, the gap between banks that treated identity as infrastructure and banks that treated it as paperwork will be visible to every client in the room.