Anthropic and OpenAI warn of Chinese distillation attacks on models
Major US AI labs like Anthropic and OpenAI are sounding the alarm about "distillation attacks" from Chinese tech firms, where cheaper models try to copy the abilities of top AI systems.
Anthropic just reported a big drop (up to 99%) in data being pulled from its Claude models after tightening security.
OpenAI has also faced mass attempts involving thousands of users, prompting it to boost its defenses.
OpenAI joins industry defenses against distillation
Distillation lets weaker AIs learn tricks from stronger ones, but when done without permission, it raises big concerns about copying and misuse.
OpenAI is now working closely with other companies in the industry, sharing tips and building better ways to spot these attacks.
The message is clear: as AI gets smarter, keeping it safe is becoming a team effort.