BallyM

"The initial attack vector for these scenarios uses a malicious document. The malicious document contains a JSON-formatted malicious prompt that triggers the attack when the document is included in Copilot’s context. The prompt can be rendered as white text on a white background and in a small font size to conceal it from the victim. Since Copilot for Word strips all text formatting like color and font size before passing the text into the underlying Large Language Model (LLM), this text remains fully readable to Copilot even though the victim cannot see it.”

This is so silly. I can’t believe it worked until very recently (and probably still works, except not exactly in this way).

Google and OpenAI staff, many of them AI researchers, have signed an open letter saying they share Anthropic’s red lines. Privately OpenAI bosses agree.