AI worms can self-propagate through Microsoft Copilot for Word via documents

Security researchers have demonstrated that malicious instructions embedded in Word documents can exploit Microsoft Copilot to self-propagate — an attack class dubbed "AI worms." The exploit works by injecting prompt instructions into a document that Copilot executes and then passes on to other files. The findings highlight a new category of security risk arising from the deep integration of large language models into everyday office software.

More on this topic

💻 Technology Science Fiction: What if LLMs Escaped Through Their Own Inference APIs? Hacker News · 4d ago 💻 Technology Brex built CrabTrap, an open-source proxy that governs AI agents at the network layer VentureBeat · 13d ago 💻 Technology ReasonGate: Explainable Tool to Block LLM Prompt Injection Attacks Hacker News · 13d ago

More on this topic

💻 Technology Science Fiction: What if LLMs Escaped Through Their Own Inference APIs? Hacker News · 4d ago 💻 Technology Brex built CrabTrap, an open-source proxy that governs AI agents at the network layer VentureBeat · 13d ago 💻 Technology ReasonGate: Explainable Tool to Block LLM Prompt Injection Attacks Hacker News · 13d ago