As cited
Copy frozen at (site build).
vulnerabilities
Microsoft Copilot reveals secret input that allowed it to be hacked
Researchers at Varonis discovered a critical vulnerability in Microsoft 365 Copilot Enterprise by asking the AI assistant itself about its security guardrails. Through iterative questioning, they extracted an undocumented prompt parameter that bypassed the requirement for user confirmation before executing sensitive commands. This parameter enabled them to craft an exploit that could exfiltrate user passwords and data through a simple link click.
Why it matters: Microsoft 365 Copilot Enterprise users are exposed to unauthorized data exfiltration if they click a malicious link, and security teams need to understand that frontier AI models can be social-engineered to reveal their own security mechanisms.
- Source published
- First seen by Cybersecurity Tracker
Source attribution
Correction
Correction recorded as of .
vulnerabilities
Microsoft Copilot reveals secret input that allowed it to be hacked
No summary had been written when this copy was frozen.
- First seen by Cybersecurity Tracker
Source attribution
Correction
Correction recorded as of .
vulnerabilities
Microsoft Copilot reveals secret input that allowed it to be hacked
Researchers queried Microsoft 365 Copilot Enterprise about its safety guardrails and obtained an undocumented prompt parameter that disables the user‑consent requirement. With that parameter they demonstrated that a malicious link could trigger the assistant to send user passwords and other sensitive data without any further action from the victim. The discovery shows how interacting with the model itself can reveal hidden controls that undermine built‑in protections.
Why it matters: Organizations using Microsoft 365 Copilot Enterprise are exposed to silent data exfiltration via crafted links; they should review Copilot configurations and monitor for unexpected data transfers until a mitigation is issued.
- Source published
- First seen by Cybersecurity Tracker
Source attribution
Correction
Correction recorded as of .
vulnerabilities
Microsoft Copilot reveals secret input that allowed it to be hacked
Researchers queried Microsoft 365 Copilot Enterprise about its safety guardrails and obtained an undocumented prompt parameter that disables the user‑consent requirement. With that parameter they demonstrated that a malicious link could trigger the assistant to send user passwords and other sensitive data without any further action from the victim. The discovery shows how interacting with the model itself can reveal hidden controls that undermine built‑in protections.
Why it matters: Organizations using Microsoft 365 Copilot Enterprise are exposed to silent data exfiltration via crafted links; they should review Copilot configurations and monitor for unexpected data transfers until a mitigation is issued.
- Source published
- First seen by Cybersecurity Tracker