History repeats itself. Remember the days of viruses that spread through Microsoft Office macros? Well, those days are back, this time courtesy of AI. Someone could send you a document with text prompts in it that, if you use for AI training, will instruct your model to do things you weren’t expecting, like corrupt your data <https://www.computerworld.com/article/4203676/copilot-worm-can-spread-through-microsoft-word-docs-2.html>. And unlike those old-style macro viruses, there is no separate “macro facility” that you can lock in a sandbox and issue warnings about: it’s all just text in the document. This quote, from Mike Leone, a VP/principal analyst at Moor Insights & Strategy, sums it up: “It’s hard not to chuckle a bit with this one. People have been asking whether data can give orders since SQL injections. We fixed that one by giving the database a way to tell an instruction from a value,” Leone said. “Thirty years later, we’ve built an entire category of software that can’t tell the difference at all.”