Skip to content
Espresso Data Privacy
EN
Request a demo

AI

Microsoft Copilot and data protection – stored AI data under control

Using Microsoft 365 Copilot creates new data relating to individuals: prompts, responses, chat and meeting transcripts. At the same time, Copilot can access everything the respective person is allowed to see in SharePoint, OneDrive, Teams and mailboxes. Data protection with Copilot therefore means keeping both under control.

What personal data does Copilot create?

Every use creates data that did not exist before:

  • Prompts and responses, often with names, cases and internal details.
  • Chat histories with the assistant.
  • Meeting recordings, transcripts and summaries.
  • Drafts and documents created by Copilot.
  • References to the content on which Copilot bases its responses.

What can Copilot see?

Copilot accesses the content that the respective person is authorised to see. That is how it should be – but it reveals what was previously hidden: documents shared with “everyone in the organisation” or “anyone with the link”, old storage locations with HR or health data, forgotten exports. What nobody used to find, Copilot finds in seconds.

Before rolling it out, it is therefore worth scanning storage locations for personal data and cleaning up content that is shared too widely.

What does data protection law require?

In 2023, the FDPIC stated that the current Data Protection Act is directly applicable to AI applications. The familiar principles apply: purpose limitation, proportionality and transparency. This leads to concrete tasks:

  • Include Copilot as a processing activity in the register.
  • Take prompts and transcripts into account for data subject access requests.
  • Define and enforce retention periods for AI data.
  • Carry out a data protection impact assessment where the risk is high (Art. 22 FADP, Art. 35 GDPR).
  • Regulate processing on behalf by the vendor in a contract (Art. 9 FADP, Art. 28 GDPR).

How long may prompts and transcripts be retained?

There is no single statutory period. What matters is the purpose: a meeting transcript that serves only as a memory aid is deleted sooner than one that counts as official minutes. Microsoft 365 offers retention settings for this. The key is to define the period, implement it – and regularly check that it is being observed.

Introducing Copilot: five steps

  1. 1Scan storage in SharePoint, OneDrive and mailboxes for personal data.
  2. 2Clean up content that is shared too widely.
  3. 3Record and classify Copilot as an AI system.
  4. 4Define retention periods for prompts and transcripts.
  5. 5Regularly check whether retention periods and vendor claims are being met.

How Espresso helps

Espresso shows which personal data Copilot can reach and what is held in stored AI data – and checks whether vendors keep their retention commitments.

  • Scan Microsoft 365

    Data Discovery scans SharePoint, OneDrive and mailboxes, including Copilot data, for personal data.

  • AI systems classified

    AI Privacy Control records AI systems, classifies them as approved, tolerated or prohibited and checks the promised retention period against the findings.

  • What Copilot can see

    Shared objects that Copilot can reach – for anyone with the link or for the entire organisation.

Frequently asked questions

Is Microsoft 365 Copilot compliant with data protection law?

That depends on how it is used. Microsoft provides settings and contracts; responsibility for purpose, permissions, retention periods and access requests remains with the organisation.

Do Copilot prompts belong in a response to an access request?

If they contain personal data of the requesting person, in principle yes – just like other stored data.

Is a data protection impact assessment needed?

Yes, if the use is likely to pose a high risk to the data subjects. This must be assessed case by case, for example when health or HR data is analysed.

Does Espresso monitor what employees enter into Copilot?

Not in real time. Data Discovery scans stored prompts and transcripts for personal data, and AI Privacy Control checks vendors' retention commitments. Espresso does not monitor data traffic; for a protected route to the AI, there is Privacy Delivery.

More topics

See Espresso in 30 minutes

We will show you the cockpit with sample data – and discuss where you should start.

Request a demo

or write directly to info@espressodataprivacy.com