OpenAI disrupts extraction attempts linked to Moonshot AI's Kimi
Summary
OpenAI reported that it thwarted a coordinated campaign to extract protected reasoning from its models, with a significant number of attempts linked to individuals associated with Moonshot AI, the developer of Kimi. This activity, which began on July 1, peaked with approximately 16,000 extraction attempts from over 4,000 users on July 24-25. OpenAI clarified that there was no breach of its databases or encrypted user conversations while it took steps to ban or restrict accounts and close the extraction pathways. The company shared its findings with industry and government partners to address similar threats within the AI ecosystem, particularly as competition intensifies among global developers.
Analysis
OpenAI: OpenAI is an artificial intelligence research and deployment company focused on developing advanced large language models and agentic systems. It recently announced new models in the GPT-6 family along with always-on agent capabilities at its annual developer event. In this news, OpenAI identified and disrupted coordinated efforts by users linked to Moonshot AI to extract protected model reasoning without any database breach. Moonshot AI: Moonshot AI is a Chinese artificial intelligence company that develops the Kimi series of multimodal models for general and specialized use cases. It has recently released open-weight versions of its latest models and conducted internal safety reviews following external testing of its systems. In this news, the company is connected through individuals to a cluster of users involved in attempting to distill reasoning from OpenAI's models. AI Security: OpenAI shared details of the extraction attempts with industry and government partners to address similar threats across the ecosystem. Model Competition: Chinese AI developers continue to advance frontier models amid global scrutiny over techniques like distillation and safety practices.
Categories
aiai_agentsmachine_learningtech
Related sources
- https://openai.com/index/introducing-gpt-6-sol-and-luna/
- https://www.theverge.com/openai
- https://openai.com/index/devday-2026-recap/
- https://www.theverge.com/ai-artificial-intelligence/1001681/openai-devday-2026-biggest-news-announcements
- https://github.com/moonshotai
- https://wccftech.com/kimi-k3-1-model-teased-within-moonshots-internal-code-snippet-and-expected-to-land-before-october-as-carnegie-finds-57-percent-of-top-global-ai-talent-now-originates-from-china/
- https://benchlm.ai/providers/moonshot-ai
- https://www.chinadaily.com.cn/a/202609/21/WS6ab0c2cbe4b06d4aa055f3fa.html
- https://eastleighvoice.co.ke/technology/406470/moonshot-ai-reviews-kimi-models-after-researchers-bypass-safety-controls
- https://openai.com/research/index/release/
- https://www.aichipmap.com/en/company/moonshot-ai/
- https://aidaily.fyi/en/t/moonshot-ai
- https://tech-insider.org/moonshot-ai-kimi-k3-bedrock-1-56tb-release-2026/
- https://www.reuters.com/business/openai-takes-meta-with-always-on-dots-agent-enterprise-ai-push-2026-09-29/