The FSB doubts humans can watch every agent step

Oversight is changing shape

The Financial Stability Board (FSB) published a consultation report setting out 12 sound practices for responsible AI adoption, and it accepts that continuous human monitoring of individual agent decisions becomes impractical as agentic AI systems multiply. It recommends supplementing human oversight with AI that monitors other AI. If you run a review queue, that is worth sitting with for a minute.

An AML agent goes on sale

FIS announced that its AML Agent, developed with support from Anthropic, is now commercially available to any U.S.-based bank, including those that are not FIS clients today. At Amalgamated Bank, a design partner, the agent is projected to reduce Level 2 case investigation time by approximately 98 percent, based on the bank's analysis of its current workflows. That figure is a projection from Amalgamated's own analysis, and it is the number to ask for again once the agent has been running for a while.

Patching is a people job too

At Bank of America, CEO Brian Moynihan told CNBC the bank put on 1,000 contractors to help with patching work in the second quarter, alongside the 3,000 it already has in that area, to deal with software vulnerabilities exposed by advanced AI models including Anthropic's Mythos. He told analysts the bank spent a "couple hundred million plus" on software remediation and regulatory issues related to anti-money laundering compliance in the third quarter, and said it expects remediation costs to decline over time. At the same time, Bank of America has about 130 internal AI projects, and Moynihan told analysts that savings from them are expected to rise from $300 million this year to $700 million in 2027.

Two places to look

Side by side, these stories put work in two places: reviewing what agents do, and cleaning up what advanced models can find. Which of the two is bigger in your own shop, and who is staffed for it?