AI poisoning could turn open models into destructive “sleeper agents,” says Anthropic
Enlarge (credit: Benj Edwards | Getty Images)
Imagine downloading an open source AI language model, and all seems well at first, but it later turns malicious. On Friday, Anthropic—the maker of ChatGPT competitor Claude—released a research paper about AI "sleeper