OpenAI says it stopped a campaign to "siphon off" its models' reasoning: the core is linked to people from Moonshot AI
Prepared by Sintelligent with the help of SI. Publication time restored from the build log. . Updated: .
Models learn from each other, and companies are fighting to keep that from becoming copying.
According to OpenAI, in the first week of July it noticed activity "resembling adversarial distillation," peaking on July 24 and 25: 16 thousand requests from more than 4 thousand users following a pattern probably meant to extract a model's hidden reasoning. The company attributes the core of the cluster to people linked to Moonshot AI (Kimi). According to OpenAI, there was no breach of databases or conversation storage.
Sources 1