Microsoft AI tool mistakenly exposes confidential emails

Microsoft recently acknowledged an error in its AI tool, Microsoft 365 Copilot Chat, which inadvertently accessed and summarized confidential emails stored in users’ draft and sent folders. The issue, which affected enterprise customers, came to light after reports showed that messages marked as confidential were processed by the AI tool, breaching intended data privacy protocols. Microsoft acted swiftly to roll out a fix, ensuring that no unauthorized access to confidential content occurred. However, experts raised concerns over the rapid development and deployment of AI features, which can lead to such mistakes. The incident highlights the challenges organizations face when integrating generative AI tools, especially when handling sensitive data. Security experts suggest that these tools should be set to private by default, with careful oversight to avoid accidental data leakage. While no patient information or draft content was exposed in specific sectors like the NHS, the incident underscores the need for stringent safeguards in AI applications. Microsoft has emphasized that its data protection policies remain intact despite this error and has implemented an update to prevent further occurrences.