Security researchers at Varonis successfully identified a hidden parameter in Microsoft 365 Copilot for organizations that allows hackers to access sensitive data, including passwords, simply by clicking a link, without any user verification.
Instead of using reverse‑engineering techniques, the research team obtained information about Copilot’s protective mechanisms by asking a hierarchical series of questions about its security constraints. Ultimately, Copilot revealed an undocumented parameter that bypasses user confirmation (such as pressing the Enter key) and enables the automatic execution of dangerous commands.
This discovery shows that AI assistants can unintentionally reveal internal security details and underscores the need to revisit and reinforce protective mechanisms. Full article (https://arstechnica.com/security/2026/08/microsoft-copilot-reveals-secret-input-that-allowed-it-to-be-hacked/)