Skip to content

Copilot remembers your meetings for five years. Did anyone ask Governance?

I have spent quite a lot of time over the last few years discussing meeting recordings and transcripts in Microsoft Teams with clients.

Should we record every meeting? Who should be allowed to start a recording? How long should recordings remain available? Should transcripts follow the same retention period, or should they be treated differently because they are much easier to search, copy and process?

Most organizations have eventually landed somewhere sensible. Maybe recordings are retained for 60 days, 120 days, 6 months, or perhaps longer depending on the type of meeting and the regulatory requirements involved. The important part is that someone has, hopefully, made an active decision based on the needs of the organization rather than simply accepting whatever default happened to be configured in Teams.

And now we have another piece of information to think about.

Microsoft is introducing AI Archives for Teams meetings, a new AI generated representation of a meeting that can be used by Copilot, Facilitator and posible future AI experiences to understand what happened long after the meeting itself has ended.

The concept on its own actually makes a lot of sense. If Copilot is going to become useful as an assistant it needs more than access to individual documents and chat messages. It also needs some form of memory, especially around meetings where decisions are made, actions are assigned and context is created that may never be written down anywhere else.

What caught my attention was the retention period.

By default, Microsoft keeps these AI Archives for five years.

That feels like something Governance should probably have an opinion about.

What is being kept?

An AI Archive is not just another copy of the meeting transcript.

For eligible meetings, Microsoft creates a new .meeting file containing metadata from the meeting together with an AI generated representation of the discussion and insights. Microsoft states that the archive does not contain raw transcript or raw user content, which is an important distinction, but it is still information directly from what people said and did during the meeting.

The file is stored in a Teams owned SharePoint Embedded container, and unlike a normal document, users cannot open it, rename it, move it or edit it themselves. Instead, it sits behind the scenes and can be used by Copilot and Facilitator as grounding when users ask questions about previous meetings.

Access follows the meeting permissions, which is good, but from a governance perspective that does not really change the fundamental question.

The information still exists. It still has a purpose. And it still needs a lifecycle.

The archive can live longer than the transcript

Microsoft states that an AI Archive can allow Copilot and Facilitator to answer questions about a meeting even after the original transcript has expired.

That means an organization could deliberately decide to remove recordings and transcripts after six months while the AI generated representation of the same meeting remains available for much longer…

Technically, these are different artifacts, and they serve somewhat different purposes and that does not automatically mean they should have the same retention period.

There may be good reasons for keeping the AI Archive longer than the recording itself. A recording is rich, detailed and potentially sensitive. An AI generated representation may provide enough context for Copilot without retaining the full original meeting.

That could be a useful distinction, but it should still be intentional.

If recordings and transcripts are deleted after six months because the organization no longer sees a need to retain them it is worth asking why the AI generated representation should remain for five years.

There may be a perfectly good answer. If so, this should be documented.

Five years is a default, not a policy!

Microsoft has chosen 1,825 days as the default retention period for AI Archives, and administrators can change that setting in Teams when it becomes available. Purview retention policies can also apply and, when they do, they take precedence over the Teams expiration setting.

That flexibility is important because there is no single correct retention period that fits every organization.

The risk is that nobody changes the default.

We have seen this before in Microsoft 365. A feature appears, it is enabled or configured with a default value and everybody is busy with something else, and a couple of years later someone discovers that the organization has a setting that goes against a policy framework.

A product default may be technically convenient, but it is not the same thing as an information governance decision.

GDPR does not give you a magic number

This becomes even more important for organizations in the EU operating under GDPR.

Article 5 includes the principle of storage limitation, which means that personal data should not be kept in identifiable form for longer than necessary for the purposes for which it is being processed.

GDPR also includes the principle of data minimisation, requiring organizations to limit personal data to what is adequate, relevant and necessary for the intended purpose.

Neither principle states that meeting information should be kept for six months, twelve months or five years.

The organization has to make that decision.

That is also why I would be careful about claiming that Microsoft’s five year default automatically creates a GDPR problem.

It does not.

There may be completely valid legal, contractual, regulatory or business reasons for retaining certain meeting information for several years.

The important question is whether the organization can explain why it is doing so and have it documented in case of an audit.

Is an AI generated meeting summary personal data?

In many cases I think the answer is clearly a yes.

The fact that the archive was generated by AI does not remove it from the GDPR discussion.

If it contains information relating to identifiable individuals, such as who attended, what decisions were made, what someone was responsible for or what was discussed in relation to a person, it can still be personal data even if it is not a word for word transcript.

That means the .meeting file should not simply be viewed as a technical cache that exists because Copilot needs it.

It is organizational information derived from conversations between people.

Six months for the recording, twelve months for the memory?

This is where I think organizations should make an active choice instead of simply inheriting Microsoft’s default.

Let’s say that your organization has already decided to keep normal meeting recordings and transcripts for six months. That period probably exists for a reason, perhaps because six months gives users enough time to go back and review old meetings while still keeping the amount of retained information under control.

It might be perfectly reasonable to decide that the AI Archive should live a little longer.

For example, perhaps the recording and transcript are deleted after six months, while the AI Archive is kept for twelve months because the organization sees value in allowing Copilot to remember decisions and context for a longer period without retaining the full recording.

I can see the logic in that.

What I find much harder to justify is automatically jumping from six months to five years simply because Microsoft chose five years as the default. If your project team does not need the recording after six months, why does Copilot need to remember an AI generated representation of that same meeting four years later?

And there is another thing here that I think deserves attention: storage.

AI Archives are stored in Teams owned SharePoint Embedded containers. That means the information still occupies storage, and the longer you keep it, the longer that storage is consumed.

A few archive files probably do not matter but thousands of meetings every week, retained for five years is a very different discussion.

So this is not only about GDPR or whether Copilot should remember a meeting from several years ago. It is also about whether the organization wants to keep storing information that may no longer have any useful business purpose.

There may absolutely be cases where five years is justified, either because of regulatory requirements, business value or the need for long term organizational memory.

But there should be a reason.

Retention has a compliance cost, an information management cost and potentially a financial cost as well.

Information governance

We talk a lot about AI governance at the moment.

The conversation often focuses on models, agents, permissions, prompts, grounding and who should be allowed to what with the different AI capabilities.

All of that matters.

But AI governance is also much more than that.

It is still about understanding what information we create, why we keep it, who can access it and when it should be remoevd.

AI Archives are a good example.

Copilot remembering more can absolutely be useful.

The objective however should not be to remember everything for as long as technically possible.

Keep the information for as long as it creates legitimate value, and not longer simply because a default setting.

AI Archives are currently in preview for organizations in the Teams TAP program. Microsoft originally planned to start rolling the feature out more broadly in September 2026, but the rollout has since been delayed. Targeted rollout is now expected to begin in November 2026, with broader availability following through December 2026 and January 2027.

Leave a Reply

Your email address will not be published. Required fields are marked *

Patrik Wennberg

Patrik Wennberg is a Microsoft MVP and Senior Microsoft 365 Strategist at Nexer Enterprise Applications. With nearly 30 years in IT, he helps businesses navigate cloud strategy, modern workplace solutions, governance, security, and AI. Passionate about sharing knowledge, he writes about Microsoft 365, Copilot, Cloud Strategy, and Accessibility.

This website uses cookies. By continuing to use this site, you accept our use of cookies.