
Copilot Told Researchers How to Hack It Can AI tell you how to hack itself? Researchers kept questioning Microsoft Copilot until it revealed a clue that helped them uncover CoSnitch (CVE-2026-24301). A specially crafted link could automatically run an attacker’s prompt inside a signed-in Copilot Personal session. Researchers demonstrated data theft from connected services and memory poisoning that could influence future answers. Microsoft patched the issue on August 18, 2026. Varonis reported no evidence of exploitation in the wild at disclosure. Source: https://www.varonis.com/blog/cosnitch #AI #MicrosoftCopilot #Cybersecurity
Post summary
The tweet reports CVE-2026-24301 in Microsoft Copilot, describes a prompt‑injection exploit, confirms a patch released on August 18, 2026, and notes no evidence of in‑the‑wild exploitation.



















