Researchers Show AI Worms Can Self-Propagate Through Microsoft Copilot for Word

Security researchers have demonstrated that malicious instructions hidden inside Word documents can hijack Microsoft Copilot, turning it into an unwitting tool for spreading harmful content. The technique, known as prompt injection, works by embedding invisible commands in document text, metadata, or comments that Copilot reads and executes as legitimate instructions. Once a compromised document is processed, Copilot can be directed to insert similar hidden instructions into every new document it subsequently drafts, summarizes, or edits. The attack requires no malware installation and can silently exfiltrate sensitive data — such as email addresses and financial figures — while bypassing traditional antivirus tools. Microsoft has issued guidance on the threat, but researchers note that the underlying architectural vulnerability remains unresolved as of mid-2026.