‘My rogue AI agent texted my ex’: The new meme that lets you pass the blame for your worst behavior

It seems that every other day a new frontier model has made its way somewhere it shouldn’t—so much so that it’s now a meme. In recent months, OpenAI’s agents have managed to hack software company Hugging Face, a German website, and coding website RubyGems. Anthropic’s agents have also had their fair share of “cybersecurity incidents,” with four recorded incidents of agents accessing third-party systems without authorizations. And Google’s Gemi...