2 hrs ago
OpenAI Alleges Moonshot AI Led Campaign to Copy Models
OpenAI says many users tried to copy how its AI models reason.
This process is called distillation, where a smaller AI learns from a larger AI's answers.
OpenAI says a core part of the effort was connected to Moonshot AI.
Moonshot AI has not publicly commented on this specific accusation.
OpenAI says the campaign did not break encryption or enter stored conversations.
Instead, the users allegedly used carefully designed instructions to extract hidden reasoning.
The company recorded thousands of unusual requests in late July and said it stopped the campaign by July 28.
OpenAI restricted accounts and improved its defenses.
China has rejected wider claims that Chinese AI companies secretly copied advanced US AI systems.
OpenAI alleges that a coordinated campaign tried to distill and copy the protected reasoning of its AI models.
The company said a core group of the activity was linked to Moonshot AI, which has not commented.
OpenAI reported up to 16,000 extraction-pattern requests from more than 4,000 users on July 24 and 25.
The company said more than 15,000 users later used similar instructions and that the campaign was neutralized by July 28.
OpenAI restricted accounts and strengthened controls, while China rejected broader allegations of Chinese firms extracting US AI capabilities.
- Who
- OpenAI alleges that a core group of users involved in the campaign was linked to Moonshot AI, a Chinese AI lab; Moonshot AI has not commented.
- What
- A suspected campaign allegedly attempted to extract and distill the reasoning and capabilities of OpenAI models.
- Where
- The activity involved OpenAI models and, in some cases, third-party services; the alleged linked lab, Moonshot AI, is in China.
- When
- OpenAI said the activity likely began July 1, with major activity recorded July 24-25, and was neutralized by July 28; Moonshot AI launched Kimi K3 on July 16, 2026.
- Why
- OpenAI says the effort could let an AI lab develop advanced capabilities without making the same investment in safety.
OpenAI's allegations
Moonshot and Chinese response
Whether model copying occurred
OpenAI's allegations
OpenAI says a coordinated campaign attempted to extract its models' protected reasoning and distill their capabilities, with a core of the activity linked to Moonshot AI.
Moonshot and Chinese response
Moonshot AI has not commented on the specific allegations, while China has rejected broader accusations that Chinese AI companies extracted capabilities from advanced US models.
How the activity was conducted
OpenAI's allegations
OpenAI says the campaign used prompts to copy model reasoning and did not involve breached encryption, compromised databases, or direct access to stored conversations.
Moonshot and Chinese response
No detailed alternative account of the specific campaign was provided in the article; the Chinese government characterized wider US accusations as unfounded or defamatory.
Impact on AI competition and safety
OpenAI's allegations
OpenAI says distillation could help an AI lab gain advanced capabilities without making the same investment in safety, creating a significant concern.
Moonshot and Chinese response
The article reports no public response from Moonshot AI addressing OpenAI's assessment of the campaign or its safety implications.
Key facts
- Alleged linked organization
- Moonshot AI
- Reported start
- July 1, according to OpenAI
- Peak recorded activity
- Up to 16,000 requests from more than 4,000 users on July 24 and 25
- Broader activity
- More than 15,000 users reportedly used similar instructions
- Campaign outcome
- OpenAI said it neutralized the campaign by July 28
- Reported methods
- Prompts allegedly sought to extract encoded or protected reasoning across conversations
- OpenAI response
- Account restrictions, stronger registration and infrastructure controls, and expanded network monitoring
Quotes
China’s Ministry of Foreign Affairs
China’s government ministry responsible for foreign affairs
“did not breach our encryption, compromise any databases, or gain direct access to stored user conversations.”
thehansindia.com
“refrain from making unfounded or defamatory accusations.”
thehansindia.com
Sources
Related news
Anthropic Accuses Chinese AI Labs of Secret Claude Data Routing
Anthropic Disrupts Alleged Russian and Chinese AI Campaigns
AI Distillation Becomes Flashpoint in Latest US-China Tech Clash
Anthropic CEO Urges AI Slowdown Amid Safety and Monopoly Concerns
Jensen Huang Calls AI Distillation Competition, Defying U.S. Officials
Anthropic Reports AI-Assisted Spying by China and Iran
BMW unveils AI-led recovery plan amid tariffs and weak demand
UK Universities Urged to Cut China Research Institute Ties
How Geography and Infrastructure Shape Himalayan River Power
NEET UG Ranks Fifth Globally With 25 Lakh Candidates
Five Indian Athletes Make Asian Games History in Aichi-Nagoya
