Microsoft Copilot Actions Features Deserve Attention

← Alle Beiträge
Künstliche Intelligenz · EN

Microsoft Copilot Actions Features Deserve Attention

With Copilot Actions, the assistant doesn’t just suggest, it acts. Booking a hotel, sending an email with local attachments, deleting duplicate files – all triggered by a prompt. These features are currently in preview for users outside the EEA, but EEA deployment is likely on the horizon.

The potential is clear: automation, productivity, and a more intuitive interface for everyday tasks. But with that comes a need for clarity – especially in regulated environments.

From a legal and privacy perspective, several areas deserve attention:

1) Controller vs. processor: When Copilot acts on the web (e.g. via Bing API), Microsoft may not operate under the DPA but as an independent controller. If this model applies to Copilot Actions, companies will need to reassess their documentation and accountability obligations under the GDPR.

2) Access to local files: Planned features include native access to local data. This raises questions about scope, consent, and technical safeguards – especially in enterprise environments.

3) Screenshot logging: Copilot Actions stores screenshots of web activity for up to 30 days. While sensitive inputs like passwords are excluded, the implications for transparency, auditability, and data minimisation must be considered.

4) Microsoft Graph access: In Microsoft 365, Copilot can access everything the user can, so emails, chats, documents, calendars. Without a strict permission model, this can lead to unintended data exposure or misuse.

5) Connected Experiences and plugins: These can lead to cross-context data processing, sometimes outside the scope of the DPA. The lack of full transparency around what data is processed and for what purpose makes it difficult to establish a reliable legal basis.

That said, there are ways to prepare:
- Start with a controlled test phase.
- Review and tighten permission concepts.
- Use Microsoft Purview to classify and protect sensitive data.
- Disable or restrict web content plugins where appropriate.
- Ensure user training on responsible AI use.

No doubt, Copilot is a powerful tool, but it changes the way we think about responsibility, access, and control. The legal frameworks are there, but they need to be applied with care.

How are you preparing for AI systems that act on your behalf? And how is this regulated in your jurisdiction?

Zu diesem Beitrag gibt es auf LinkedIn ein Bild. Auf LinkedIn ansehen →

Auf LinkedIn diskutieren