In mid-July 2026, an old Anthropic blog post resurfaced, accusing Moonshot AI of distilling Claude through 3.4 million fraudulent interactions. The timing coincided with the release of Kimi K3, amplifying the controversy. OpenAI's Dean W. Ball suggested the US government create regulatory risks for Chinese open-source models without evidence, but quickly distanced himself from the advocacy.
Meanwhile, on July 20, a paper titled 'Cura 1T: Specialized Model for Agentic Healthcare' appeared on arXiv, authored by US medical AI startup actAVA. The paper explicitly states that Cura 1T is post-trained on Kimi-K2.6, an open-weight model from Moonshot AI. The team, including ex-Salesforce engineers, used a rank-32 LoRA adapter to fine-tune the model, offering it as a closed-source API at $0.5/M input tokens and $2.5/M output tokens—half the price of Kimi's official API. They claim Cura 1T outperforms GPT-5.5, Claude Opus 4.8, and Gemini 3.1 Pro on 5 out of 6 medical benchmarks.
This is not an isolated case. In March, Cursor was caught using Kimi K2.5 as the base for its Composer 2 model, initially omitting the attribution. After public scrutiny, Cursor acknowledged the base model and later included it in their technical report. By July, using Kimi as a base has become normalized, with actAVA openly declaring it in their paper and website.
The contrast between the narrative of theft and the reality of US companies adopting Chinese models is stark. While Anthropic and OpenAI push accusations, US enterprises like Cursor, actAVA, Cloudflare, and Airbnb are integrating Kimi and Qwen into production. Kimi-K2.6 saw over 1.15 million downloads on HuggingFace in the past month alone. The 'distillation' narrative may be losing ground as the industry votes with its orders.
Key upcoming events include the full release of Kimi K3 weights by July 27, which could trigger a wave of derivative models. The Cura 1T API is open for third-party verification, and policy responses to Chinese open-source models are expected within two weeks. The shift in attitude from apology to open adoption marks a significant change in the AI landscape.