---
title: "The evidence trail: what a supervisor will ask for after an incident"
type: Article
published: 2026-11-09T06:00:00.000Z
url: https://agenticopenfinance.com/blog/week-2026-11-09
source: AgenticOpenFinance
---

# The evidence trail: what a supervisor will ask for after an incident

> After an agent incident, supervisors ask who authorised what, what the agent did and how the firm knew. Logging, documentation and evidence packs answer those questions.

When an agent causes harm, the investigation follows a predictable path. A supervisor, an auditor or a court will ask: **who authorised this, what exactly happened, and how did the firm know and respond?** The firm that can answer quickly, with records it did not have to reconstruct, is in a far better position.

**What the rules already expect.** For high-risk AI systems, the EU AI Act requires automatic logging over the system's lifetime, technical documentation kept up to date, post-market monitoring, and reporting of serious incidents to authorities. Deployers must keep the logs under their control for a minimum period. In banking, the UK PRA's model risk principles and the US agencies' revised model risk guidance expect inventories, documentation and monitoring, and the US guidance explicitly leaves generative and agentic AI to firms' broader risk management.

**What an agent's record should contain.** An [evidence pack](/glossary/evidence-pack) for each action:
- the agent, its version and its operator;
- the principal and how they authenticated;
- the mandate and the limits it set;
- inputs, each tool call and its result;
- approvals, overrides and timestamps.

**Making it trustworthy.** Records kept by the party under investigation carry more weight when they are [tamper-evident](/glossary/tamper-evident-log), for example hash-chained with signed checkpoints, and when retention follows a documented schedule.

The test is simple: could someone outside the firm verify what happened without taking the agent's word for it?

**Sources:** [EU AI Act, Article 12: Record-keeping (AI Act Service Desk)](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-12), [EU AI Act, Article 72: Post-market monitoring (AI Act Service Desk)](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-72), [EU AI Act, Article 73: Reporting of serious incidents (AI Act Service Desk)](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-73), [PRA SS1/23: Model risk management principles for banks](https://www.bankofengland.co.uk/prudential-regulation/publication/2023/may/model-risk-management-principles-for-banks-ss), [Revised Guidance on Model Risk Management (Fed, OCC, FDIC), SR 26-2 attachment](https://www.federalreserve.gov/supervisionreg/srletters/SR2602a1.pdf)
