Nouvelle panne d'IA chez Microsoft : Copilot révèle des informations secrètes

A recent vulnerability in Microsoft Copilot has exposed confidential information, highlighting ongoing challenges in securing advanced AI tools. The issue, stemming from a programming error, allowed the chatbot to display sensitive emails and files, particularly from the "Sent Items" folder, even when protected by security labels. This incident underscores the critical need for robust security measures as AI technologies become more integrated into daily workflows. The flaw primarily affected the "Sent Items" folder within Microsoft's ecosystem, meaning confidential communications could have been inadvertently revealed. While Microsoft is reportedly deploying a patch to rectify the vulnerability, the incident raises questions about the thoroughness of security testing for AI-powered features. The precise technical details of the bug have not been fully disclosed, but its impact points to a failure in access control mechanisms within Copilot's processing of user data. This security lapse has significant implications for businesses and individuals relying on AI assistants for productivity and communication. Users of Microsoft Copilot, especially those handling sensitive data, are now more aware of potential risks associated with the technology. The incident also fuels the ongoing European debate around digital sovereignty and the reliance on non-European tech providers for critical infrastructure and services, prompting calls for greater scrutiny and development of local AI solutions.