AI models like OpenAI’s ChatGPT and Google’s Gemini can be “poisoned” by inserting just a tiny sample of corrupted documents into their training data, researchers have warned. A joint study between ...