What is an audit trail?
An audit trail is a chronological record of who did what and when, kept so you can reconstruct how something reached its current state. A version history where every version is kept with the name of whoever made it is an audit trail. The record is only useful if it cannot be edited quietly and nothing is skipped.
Part of Glossary: the terms behind MCP, coding agents and shared pages, defined plainly.
Every action that matters writes one entry: who (a person, or an agent acting for them), what changed, which thing, and when.
Entries are appended, never edited or deleted. If a change is undone, the undo is a new entry.
Each entry carries enough to reconstruct the event: the version before and after, or a diff, not just a label.
Later, someone reads the entries in order to answer: how did this get here, who approved it, and what did the agent do between the two human steps.
Source: NIST Computer Security Resource Center glossary: audit trail
Audit trail versus a log
A log is whatever a system writes down as it runs: requests, errors, timings, debug noise. An audit trail is a narrower thing with a stricter job. NIST defines it as a chronological record that lets you reconstruct and examine the sequence of activities leading to an operation or event, from inception to result. The three words that matter are chronological, reconstruct and sequence. A trail is complete for the actions it covers, it is in order, and each entry carries enough that you can replay what happened without the people who did it in the room.
So a log can be sampled, rotated and lost. An audit trail cannot, or it stops being one. And a log is for the operator while a trail is for whoever has to answer the question afterwards: a reviewer, a customer, a regulator, or you in six months.
Example: a page edited by people and agents
A team keeps a planning page that four people and their coding agents edit. Each publish becomes a new version, and the version records who made it, when, and what changed against the previous one. Version 7 was Ana at 09:12; version 8 was Ben's agent at 09:40, touching only the pricing table; version 9 was Ana again, reverting one row. Nothing is overwritten. Read the versions in order and you can see how the page reached its current state, which edits were a human's and which an agent's, and exactly where the pricing number changed. That version history, every version kept with who made it, is an audit trail with no extra machinery. Intraface boards work this way.
Why it matters
Agents change the math. A person makes a few changes an hour; an agent can make dozens in a minute, each of them plausible. You will not review them as they happen. The trail is how you review them afterwards, find the one that introduced the wrong number, and see which instruction caused it. Without it, trust in agent work comes down to vibes.
It also settles disputes without blame. When who did what is a fact in a table rather than a memory, the conversation moves from who broke it to how we stop it breaking.
Common mistakes
- Editable entries. If an admin can fix a trail entry, the trail proves nothing. Append only; corrections are new entries that point at the old one.
- Recording the actor as the tool. An entry that says changed by bot or by the API has lost the one thing it was for. Agents act for people; record the person, and the agent as a detail.
- Keeping the what but not the before. A label like updated pricing is not reconstructable. Keep the previous version, or a diff, so the change itself can be examined.
- Trusting timestamps from the client. Order and time should be assigned where the entry is written, not by whoever submits it.
- Deleting old versions to save space. The oldest entries are the ones you need when something has been wrong for a long time.
FAQ
- Is an audit trail the same as a log?
- No. A log is anything a system writes as it runs and can be sampled or rotated. An audit trail is a complete, ordered, append-only record of who did what and when, kept so events can be reconstructed later.
- Why does an audit trail matter when AI agents edit things?
- Agents make many changes quickly and you will not watch each one. The trail lets you review afterwards: which version an agent produced, from which instruction, and where a human stepped in. Keeping every version with its author is the simplest way to get one.
- Does an audit trail have to be tamper-proof?
- In practice, yes. If entries can be edited or removed quietly, the record cannot be trusted to reconstruct anything. Append only, assign time and order where the entry is written, and keep the old versions.
Stop sending HTMLover Slack.
Create an account in a minute, connect your coding agent, and make your first board.