Three pieces this week. Each one hands the agent record to a different party at a different moment. The second line, the carrier, the buyer. None of them inherits the record by default.
This week.
Where Model Risk Ends (May 25). Model risk management was built for an artifact that does not move. An agent run is an artifact that does. SR 26-2 carved agentic AI out of model-risk scope, and the run still has to live somewhere. The handoff is where the record goes missing.
The Underwriting Surface(May 27). An insurance market for adaptive-AI errors is forming. The product needs an underwriting surface, and the agent run does not present one today. Carriers writing AI E&O are pricing without loss triangles. Reinsurers are quoting without a forensics path. Neither position holds.
The Procurement Clause (May 29). SR 26-2 told institutions to govern agentic AI with existing risk practices written for a model you buy once. An agent run is a service you rent, and the record lives wherever the vendor decides. You have leverage to require the record exactly once. Before signature.
What changed.
The EU moved this week. The Council and Parliament reached political agreement on May 7 on the Digital Omnibus package, a streamlining of the AI Act. Some Annex III high-risk obligations that were due to enter into force on August 2, 2026 now shift toward December 2027. The deadline pressure eased. The substance did not. A high-risk agentic deployment still has to produce evidence when the obligation lands, and a later date is more runway, not a smaller record.
SR 26-2 still carves agentic AI out of scope pending the interagency RFI. No new movement on that front in the last seven days. The agencies have signaled the RFI will address generative and agentic AI directly. That is the surface where the federal gap gets named in writing.
The NAIC AI Systems Evaluation Tool pilot runs through September. Twelve states are participating. The tool gets revised on pilot feedback in the fall and is expected to be re-exposed ahead of the November meeting. The state insurance examiner is still on track to move before the federal banking examiner does.
What we are tracking.
The AI E&O market. If carriers keep writing adaptive-AI coverage without a replayable record, the first contested claim sets the precedent for what the policy actually covers. The underwriting surface gets defined in a dispute or it gets defined in the contract. One of those is cheaper.
The interagency RFI. The agentic carve-out in SR 26-2 is a deferral, not a resolution. When the RFI opens, the question of what an agent run has to retain becomes a comment-period question, and the answer gets read back into examination practice.
The procurement language. Buyers are still encoding the decision record into contract terms. Every RFP that asks how the vendor signs and exports the record moves the artifact from a nice-to-have into a condition of sale.
Thirty minutes. Architecture, not sales.
A conversation about what the trajectory record has to contain to survive the next examination cycle, and where the artifact should live in your organization.
JonathanLuethke@WayfinderSystemsGroup.com