US security agencies tell AI firms to degrade answers to suspected Chinese distillation traffic in secret

The NSA, CISA, and the FBI are asking American frontier AI developers to treat model outputs themselves as an exfiltration surface, in a joint Cybersecurity Advisory that names six China-based AI companies as running what the agencies call industrial-scale knowledge distillation campaigns against US models. The most operationally striking recommendation is not a block or a ban: the agencies want providers to subtly alter responses to high-confidence malicious distillation requests, without telling the user.