OpenAI Astra distillation is drawing attention after a report alleged that Chinese AI companies DeepSeek and Moonshot AI used outputs from OpenAI’s GPT-6 Astra to train their latest models. The report highlights growing tensions over how AI developers use competing models to improve their own systems.
According to a report by The Information, cited by Reuters on October 29, both companies relied heavily on Astra’s outputs during model development. The report cites people with direct knowledge of the matter. The allegations come as American AI companies increase efforts to protect their models from unauthorised extraction of their capabilities.
OpenAI Astra distillation: What the report reveals
The report identifies two AI models that allegedly benefited from distillation using OpenAI’s Astra.
Moonshot AI reportedly used Astra outputs while developing Kimi K3.1 Pro. DeepSeek’s V4.1-Flash 1017, released on October 17, also reportedly benefited from the technique.
However, the report does not establish that distillation was the sole reason for these models’ progress. Research, computing resources and training methods also influence AI performance.
OpenAI, DeepSeek and Moonshot AI did not immediately respond to Reuters’ requests for comment, according to the report.
What is AI distillation?
AI distillation is a training technique that helps one model learn from another model’s outputs. Developers can use these outputs to build training datasets and improve performance on specific tasks.
The technique is not inherently improper. AI companies also use distillation internally to create smaller, faster and less expensive models.
The controversy arises when developers extract capabilities from proprietary models in ways that violate the provider’s terms or restrictions. Such practices can raise questions about intellectual property, access controls and development costs.
For companies building advanced AI systems, unauthorised extraction could allow competitors to reproduce certain capabilities without making comparable investments in research and computing infrastructure.
OpenAI Astra distillation raises security concerns
OpenAI has previously raised concerns about attempts to extract capabilities from its AI systems.
In a September 30 blog post, OpenAI said it had linked a core part of a distillation campaign that began in July to individuals associated with Moonshot AI. This earlier statement is separate from the specific allegations in the latest report.
OpenAI has said it takes model distillation concerns seriously. The company continues to develop safeguards against attempts to extract its models’ capabilities.
Anthropic has also introduced measures to prevent competitors from harvesting outputs from its Claude models. In October, the company described efforts to limit data collection through suspected distillation attacks.
These developments show why AI developers are increasingly focused on protecting their proprietary models and training methods.
Why AI distillation matters in the US-China technology race
The reported use of Astra outputs reflects broader competition between American and Chinese AI developers.
Companies in both countries want to build more capable models while controlling training and operating costs. Distillation can help developers improve performance without repeating every stage of training a larger model from scratch.
However, distillation is only one part of AI development. Research, computing hardware, data quality and model architecture also influence results.
The issue has become particularly sensitive as leading AI companies seek to protect their technological advantages. Restrictions on advanced models and computing resources have added another dimension to the competition.
What happens next?
The OpenAI Astra distillation allegations raise questions about how developers should protect proprietary models while allowing legitimate research and commercial use.
OpenAI and other AI companies may strengthen safeguards against large-scale extraction attempts. These measures could include better detection systems, usage restrictions and tighter access controls.
For the wider AI industry, the challenge is to distinguish legitimate model training from practices that violate access policies or undermine developers’ investments.
The allegations involving DeepSeek and Moonshot AI underline how important that distinction has become. As competition intensifies, AI distillation is likely to remain a major issue for companies seeking to protect their models and maintain a competitive advantage.
Read more : AI agent safety monitors
