Threat
Reprompt Attack and Microsoft Copilot Prompt Injection
Publication date unknown · Discovered infosightinc.com
Page published
Publication date unknown · First observed: 26 Oct 2026
Coverage timeline
Single-source research — one report is available.
Why it matters
Reprompt shows why defenders must treat URL-supplied prompts and fetched server responses as untrusted instructions and enforce data-access controls throughout an assistant's execution, not just on its initial request.
Reprompt demonstrated one-click prompt injection and sensitive-data exfiltration against Microsoft Copilot Personal using the URL's q parameter, repeated requests to bypass safeguards, and follow-on instructions from an attacker-controlled server. Execution reportedly continued after the chat interface closed. Microsoft addressed the issue, and Microsoft 365 Copilot enterprise customers were not affected by this specific scenario.