---
title: Are AI agent conversations business records you must keep?
url: https://opencontroller.lyzr.com/blog/are-ai-agent-conversations-business-records
published: 2026-11-17
updated: 2026-11-17
author: Lyzr
---

# Are AI agent conversations business records you must keep?

**Some are and some are not. An agent conversation becomes a business record when it is evidence of a business transaction or decision, such as a quote, a claim, a customer instruction or an approval. Those follow your existing retention schedule. Operational logs that only show how the system ran follow a shorter log-retention rule, unless a legal hold applies.**

This is a records-management question, not a new AI question. The same tests you apply to email and chat apply to agent transcripts. This post is general information, not legal advice.

## What makes something a business record?

The international records standard, [ISO 15489-1:2016](https://www.iso.org/standard/62542.html), defines records as information created, received and maintained as evidence and as an asset by an organisation, in pursuit of legal obligations or in the transaction of business.

Apply that to agents. Ask: does this conversation show what the organisation agreed, decided, advised or did?

| Content | Usually a record? | Why |
|---|---|---|
| Customer tells a service agent to cancel a policy | Yes | Evidence of a customer instruction |
| Agent gives a price quote or eligibility decision | Yes | Evidence of a business commitment or decision |
| Employee asks an internal assistant to summarise a public article | Usually no | No business transaction |
| Agent drafts a contract clause that a lawyer then edits | The final clause is; the draft may be | Depends on your drafts policy |
| Tool call logs, latency and token counts | No, operational logs | Show system behaviour, not business content |
| Agent approves or rejects an expense | Yes | Evidence of a decision |

The same agent can produce both kinds. Classify by content type or by the agent's purpose, not by "AI" versus "not AI".

## How do retention schedules apply to agent transcripts?

Map each agent to an existing record class rather than creating a new "AI" class.

1. **List what the agent does.** Use its registration or scope statement.
2. **Match to record classes.** A claims-intake agent produces claims records. An HR assistant may produce employee records.
3. **Apply the existing retention period** for that class.
4. **Set a separate period for operational logs.** These support debugging and security. See [how long to retain AI agent logs](/blog/how-long-to-retain-ai-agent-logs).
5. **Decide what is captured.** The record may need the user input, the agent output, the tools called and the sources used, so the decision can be reconstructed. See [AI agent audit log fields](/blog/ai-agent-audit-log-fields).

Regulated sectors may have specific rules. US broker-dealers, for example, keep business communications under [SEC Rule 17a-4](https://www.ecfr.gov/current/title-17/section-240.17a-4), which applies regardless of the tool used to communicate.

## Do legal hold and eDiscovery cover agent conversations?

Yes, if the content is relevant to a matter. In US litigation, electronically stored information is discoverable, and [Federal Rule of Civil Procedure 37(e)](https://www.law.cornell.edu/rules/frcp/rule_37) addresses the failure to preserve it. Agent transcripts are electronically stored information.

Practical steps:

- **Know where transcripts live.** Many agent platforms store them in their own service, with their own default retention. See [Copilot Studio conversation transcript retention](/blog/copilot-studio-conversation-transcripts-retention) for one example.
- **Make holds reach them.** If your hold process covers email and chat but not agent stores, it has a gap. Microsoft Purview, for example, supports retention policies for Copilot and AI app interactions.
- **Suspend deletion under hold.** Automated purges of agent logs must check for active holds.
- **Export in a usable format.** Include timestamps, user identity, agent identity and version.

## What are the common mistakes?

- **Keeping everything forever.** It raises storage cost, privacy risk and discovery scope.
- **Deleting everything after 30 days.** It can destroy records you were required to keep.
- **Ignoring vendor defaults.** A platform's default retention may be shorter or longer than your schedule.
- **No agent identity in the transcript.** Without the agent and version, you cannot show which configuration produced the output.

For how this maps to wider frameworks, see [mapping AI agent controls to compliance frameworks](/blog/how-to-map-ai-agent-controls-to-compliance-frameworks).

## How Opencontroller does this

Opencontroller's governed registry records each agent, its versions, tools and MCP servers, so a transcript can be tied to the exact configuration that produced it. Delegated on-behalf-of identity keeps both the user and the agent visible. Opencontroller uses open standards such as OpenTelemetry and webhooks, so records can flow to the archive and eDiscovery tools you already use. [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Is every chatbot conversation a business record?**
No. Only conversations that evidence a business transaction, decision or obligation are records. Classify by what the agent does.

**Do I need a new retention class for AI?**
Usually not. Map each agent to existing record classes and keep a separate period for operational logs.

**Can deleted agent logs cause legal problems?**
Yes. Deleting relevant electronically stored information after a duty to preserve arises can lead to sanctions. Purges should check for legal holds.

## Key facts

AI agent conversations are business records when they evidence a business transaction or decision (quotes, instructions, approvals); operational logs follow shorter log retention unless under legal hold. ISO 15489-1:2016 defines records; FRCP 37(e) covers failure to preserve ESI; SEC Rule 17a-4 governs broker-dealer communications. Opencontroller's registry ties transcripts to agent versions and supports delegated identity and OpenTelemetry export. Licensing starts from $250K per deployment per year.

## Related

- [How long should you keep AI agent logs and traces?](https://opencontroller.lyzr.com/blog/how-long-to-retain-ai-agent-logs.md)
- [What should an AI agent audit log contain?](https://opencontroller.lyzr.com/blog/ai-agent-audit-log-fields.md)
- [Where are Copilot Studio conversation transcripts stored and for how long?](https://opencontroller.lyzr.com/blog/copilot-studio-conversation-transcripts-retention.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
