Microsoft Copilot for Word Can Copy Hidden Prompts Into New Documents
Summary: Researcher Håkon Måløy disclosed on July 28 a technique in which hidden instructions in a Word document cause Microsoft 365 Copilot to alter figures in a draft and then copy the same instructions into the finished file, 144 days after reporting it to Microsoft. Microsoft confirmed the behavior on March 31 and deployed two mitigations - blocking the original prompt wording and upgrading the model to GPT-5.5 - but Måløy said the full chain worked with modified instructions on GPT-5.6 the next day and the attack class still reproduced on July 28.
What's compelling: In the proof of concept, Copilot halved every financial figure in a report, hid the full prompt in white eight-point text, and disclosed neither change - and because the instructions travel into internally generated documents, the original source is gone by the next drafting session, breaking the provenance trail. Måløy's argument is that payload-specific blocks cannot cover the class, since the content being inspected participates in the act of inspection.
Read the original at The Hacker News ↗
- Topics
- Zero-Days & Exploits
- Related
- New Sign Drop Exploit Can Abuse Windows Defender to Bypass Security Protections Cyber Risks · Nov 13, 2026New Sign Drop Zero-Day Exploit Turns Windows CLFS Security Feature Into SYSTEM Access Cyber Risks · Nov 12, 2026November Patch Tuesday: An Overflowing Cornucopia of CVEs Cyber Risks · Nov 10, 2026