Researcher identifies prompt injection vulnerability in Microsoft 365 Copilot for Word

The record

Written from the 1 report below. Nothing here is unsourced.

A researcher, Håkon Måløy, disclosed a persistent prompt injection vulnerability in Microsoft 365 Copilot for Word that lets hidden instructions in a document alter Copilot's output—such as halving financial figures—and then copy those instructions into the generated file to spread further.
Microsoft deployed two mitigations, including blocking the original wording and upgrading the model, but Måløy says the attack class still worked on the day of publication.
The attack is not zero-click and requires a Copilot drafting or editing operation where the malicious document enters the model's context.
No exploitation in the wild has been reported, and no public CVE or standalone Microsoft advisory exists for the finding.

Whether Microsoft issues a public CVE or advisory for the Word finding.
Whether the exploit class remains workable against future model upgrades given Måløy's GPT-5.6 workaround.
Whether detection coverage for this payload in Defender for Office 365 or Copilot's runtime safeguards is confirmed.

1 report

English national1

All filed from India

Named United States · Microsoft · Microsoft 365 Copilot · Word · Håkon Måløy

The 1 report is listed beside the record.

Answers cite the reports above, or say they can't.