OpenAI claims Chinese model stole its signature IP, which it originally lifted from others
openai
| Source: Mastodon | Original article
OpenAI alleges a Chinese AI model has copied its proprietary technology, despite the company’s own reliance on widely used IP.
OpenAI has announced that it disrupted an “adversarial model distillation” campaign run by Chinese start‑up Moonshot AI. In a blog post on Wednesday, the company said it detected a coordinated effort to extract proprietary reasoning capabilities from its models between 1 July and 28 July. The attack peaked on 24‑25 July, when more than 16 000 requests were logged from over 4 000 distinct users, and the overall investigation flagged suspicious activity linked to more than 15 000 accounts.
The claim adds OpenAI to a growing list of U.S. AI firms and government officials who allege that Chinese rivals are using large‑scale data scraping and distillation to replicate the performance of American models. Google, Anthropic and other industry players have voiced similar concerns, and the U.S. administration has framed such practices as a “national security risk.” OpenAI’s statement underscores the emerging threat of intellectual‑property theft that does not involve physical supply‑chain disruption but instead targets the software layer of advanced models.
The episode matters because it highlights a new frontier in AI competition: the ability to “distill” a model’s capabilities into a separate system without access to the original code or weights. If successful, such attacks could erode the commercial advantage of leading developers and raise broader safety questions, as the stolen reasoning patterns may be repurposed without the original safety controls. Anthropic’s recent release of Claude Opus 5.5, which includes a “preserved thinking” defense against distillation, illustrates how vendors are beginning to embed technical safeguards.
Watchers should monitor whether OpenAI pursues legal or regulatory action against Moonshot AI, how other firms reinforce their models against similar attacks, and whether policymakers introduce stricter rules on cross‑border model scraping. The incident may also accelerate industry collaboration on detection tools, echoing earlier warnings from leaders like Satya Nadella that all AI models should be assumed compromised until proven otherwise.
Sources
Back to AIPULSEN