Microsoft's AI Copilot Oops: Confidential Emails Peeked by Mistake!

Microsoft recently admitted to a mistake involving its AI assistant, Copilot, which is used in Microsoft 365 programs like Outlook and Teams. The company designed Copilot to help workers by answering questions and summarizing emails, but a recent error caused it to access and display confidential emails from users' Drafts and Sent folders. Even emails marked as confidential were included in these summaries, raising concerns about privacy and data protection. Microsoft responded by releasing an update to fix the issue and assured users that no one was given access to information they were not already allowed to see. However, some technology experts believe that as companies rush to add new AI features, mistakes like this are bound to happen.

The problem was first reported by a technology news outlet, which saw a service alert confirming the issue. The alert explained that emails with a confidential label were being processed incorrectly by Copilot Chat. This meant that even when companies had set up rules to prevent sensitive information from being shared, Copilot could still summarize and show these emails. Microsoft said that their security controls were still in place, but the way Copilot worked did not match their intentions. The company quickly sent out a configuration update to all enterprise customers to solve the problem.

Experts say that using AI tools in the workplace can be risky, especially when new features are released quickly. Nader Henein, a data protection and AI governance analyst, said that mistakes like this are almost impossible to avoid because AI technology is changing so fast. He explained that organizations often do not have the right tools to protect themselves when new features are added. Another expert, Professor Alan Woodward, said that AI tools should be private by default and only used if people choose to turn them on. He warned that even if companies do not mean to share private information, accidents can still happen. This incident shows that while AI can be helpful, it is important to be careful and make sure privacy is protected.