Leveraging Microsoft Copilot for Credential Access

Bottom line up front

In a recent LinkedIn post, I highlighted a potential security concern with Microsoft Copilot: its ability to surface sensitive credentials stored across internal platforms like SharePoint and OneDrive.

AI is Here. Is Your Security Posture Ready?

Copilot isn’t malicious, but it’s effective at retrieving data based on prompts like:

"Find all documents that contain plaintext passwords"

This functionality can unintentionally (or intentionally) expose secrets, credentials, API keys, database URLs if they’re stored without proper controls. It’s not an external breach, but it does qualify as Credential Access under tactics like those tracked in MITRE ATT&CK.

Following my post, several colleagues have shared that this issue is already being discussed in their organizations which is always a good sign for security awareness. The conversation is shifting from “how AI helps productivity” to “how AI magnifies existing security hygiene problems.”

This also ties into Nabil Aitoumeziane’s post, which explores how user prompts can drive sensitive data exposure. When combined with permissive file access, Copilot can 'show you where the risks already are'.

Nabil also references a great article written by Roman Avanesyan to support this idea.

What Security Teams Should Do

• Enforce password management policies. Plaintext passwords should never be stored in documents, Copilot or not. Use secure vaults and credential managers organization wide.

• Audit internal storage (SharePoint, OneDrive, Teams, etc.) for secrets and hardcoded credentials using automated scanners.

• Review and restrict file access controls. Limit who can access sensitive documents and ensure least privilege principles are applied across M365 resources.

• Set guardrails for Copilot usage. Not all users need Copilot access to all data. Define usage policies and scope what Copilot can search.

• Educate staff on prompt hygiene. Help employees understand how even seemingly harmless prompts can reveal sensitive internal information.

• Stay informed on emerging AI trends. As AI adoption accelerates, the threat landscape is evolving just as quickly. Staying current is critical for anticipating new risks and adapting your security posture.

Conclusion

AI tools like Copilot don’t create new threats. They expose existing ones with unprecedented efficiency. It’s up to security teams to catch up before threat actors and red teamers inevitably take advantage.

Further exploration of everyday use tools for Red Teaming

Check out how Anthony Fu leveraged DocuSign to catch voluteered credentials.