OpenAI says it disrupted a coordinated July 2026 campaign that tried to copy the behavior and reasoning of its AI models at scale, allegedly involving users linked to China’s Moonshot AI. According to OpenAI, the activity began July 1, spiked on July 24-25 with about 16,000 extraction-pattern requests from more than 4,000 users, and showed related prompt activity across more than 15,000 accounts. OpenAI said there was no database breach or direct access to stored chats, but attackers abused model interactions to reproduce protected reasoning visible to requesters, then forced the campaign offline on July 28 and closed a replay path for encrypted reasoning.
Why it matters: This is a security story because it describes a large-scale extraction attack against a major AI provider, with implications for model theft, safeguard bypass, and AI supply-chain trust. Organizations using frontier AI services should watch for vendor guidance on abuse controls and understand that attacks may target model outputs and reasoning rather than traditional data stores.
2026.09.30
100% relevant
This article appears to be the first tracked item centered on OpenAI's disclosure of the July 2026 reasoning-extraction and distillation campaign allegedly tied to Moonshot AI.
← Back to all stories