The OCC's first AI framework asks the right questions
Three weeks after JPMorgan, Citi and Bank of America started restricting outside AI agents, a US banking regulator has put its thinking on paper. On November 4, the Office of the Comptroller of the Currency (OCC) issued a supervisory framework for examiners assessing how national banks use AI, which it now calls “Super Intelligence” following an executive order. It's guidance for examiners, not new rules for banks, and most of it covers banks' own use of AI. But it also addresses the agents banks don't control, and what it says about them is worth reading.
Agents the bank doesn't control
The 22-page framework treats AI agents run by customers, counterparties, fintechs and agentic-commerce companies as channel risk: systems that act through a bank's channels without being the bank's to manage. It points examiners to identity, provenance, delegation, authentication, monitoring, velocity limits, dispute handling, suspension and revocation, calibrated to what the agent is trying to do. And it judges risk by consequences rather than sophistication. A fully autonomous, customer-facing payment agent sits in the highest risk tier, regardless of how well it performs in testing.
The risks that sit between firms
Several of the risks the OCC names can't be seen from inside a single bank. It lists opaque provider chains and concentration in common models and infrastructure among its illustrative third-party risks, and asks examiners to consider correlated or cascading activity across customers, agents and providers. It also flags point-in-time validation as a weakness, because vendor and model changes can alter behavior between periodic reviews. Each of those is a question about the network a bank sits in, as much as about the bank itself.
The question it leaves open
What the framework doesn't address is who pays the customer when an outside agent gets it wrong. It tells banks to consider how their contracts with vendors assign responsibility for errors and losses, and it is clear that accountability for the bank's own risk stays with the bank. It says nothing about the customer whose agent moves money they didn't intend to move. Banks are answering that themselves: from November 24, Wells Fargo's updated terms treat anything an agent does with a customer's credentials as authorized by the customer. That's the liability transfer Jamie Dimon warned about, and the federal question of whether consumer protections still apply when a customer has handed an agent their password remains unanswered.
The OCC has started in the right place: knowing who stands behind an agent, watching how risk changes rather than checking it once, and looking for concentration across providers. Doing that for every company in the chain, and agreeing in advance where liability lands when something breaks, is work no single bank can do on its own. It's the infrastructure layer Invela is building.
Invela is the infrastructure layer that makes open finance trustworthy - accrediting who's in the network, monitoring risk in real time, and ensuring liability lands in the right place.
