Skip to content
Submit incident
AII-1001-11747
ConfirmedA court, a regulator or the organization itself has stated the core facts, or the record cites a court or regulator source.Supporting sources are pending verification.

Microsoft Copilot read confidential emails despite privacy labels

Microsoft's Copilot Chat AI read confidential emails in January 2026 despite privacy labels meant to prevent analysis. The company confirmed the bug and deployed a fix, stating no unauthorised third-party disclosure occurred, but the incident raised concerns about privacy controls in AI systems with access to sensitive user data.

January 2026 Β· Microsoft Β· Privacy and surveillance
January 2026
AII-1001-11747
Quick facts
Category
Privacy and surveillance
Country
Slovakia
Date
January 2026
Drafted by
Raidu Incident Detection

What happened

In January 2026, Copilot Chat – an AI chat function in Microsoft 365 applications including Outlook, Word and Teams – was found to process emails marked as confidential or protected by labels intended to prevent analysis by autonomous tools. The AI read messages from Sent and Drafts folders and created summaries in responses to user questions despite the privacy markings.

Reported impact

The incidents raised concerns among security experts and IT administrators. Reports indicated AI assistants could access large volumes of user data including business, medical, legal and strategic secrets, creating risks of privacy violations and legal conflicts when AI systems could read confidential data.

Index analysis

Microsoft confirmed the bug and began deploying a fix. The company stated it was not an unauthorised disclosure to third parties and the system did not provide access beyond user permissions. However, the ability of AI to process emails marked confidential despite user protections indicated a failure in the implementation of privacy controls designed into the system.

Governance controls

AI tools like Copilot are intended to function as intelligent assistants that summarize emails, create draft responses and organize information from documents. In practice this requires AI to access large quantities of user data, often including confidential information, creating tension between functionality and privacy protection.

Classification

Category
Privacy and surveillance
Industry
Digital security
Country
Slovakia
Harm status
Actual harm
Intent
Unintentional
Timing
Post-deployment
Harmed
Business, Consumers
Failure modes
Agent misaction, Unsafe autonomous behaviour
AI systems
Copilot
Regulations
GDPR (cited)

Sources and evidence

Sources not yet published

This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.

How sources are verifiedSuggest a source for this incident