DeepSeek and Moonshot AI Allegedly Trained Latest Models Using OpenAI's GPT-6 Astra

Chinese AI Labs Accused of Using GPT-6 Astra for Model Training

Chinese AI developers DeepSeek and Moonshot AI relied heavily on outputs from OpenAI's GPT-6 Astra to train their latest models, The Information reported on Wednesday, citing people with direct knowledge of the matter1

1

2

2

. The models in question are Moonshot AI's Kimi K3.1 Pro, released last week, and DeepSeek's upgraded V4.1-Flash 1017 model, which launched on October 173

3

. This revelation has reignited debate over AI distillation practices and their role in US-China tech competition.

Understanding AI Distillation and Its Competitive Implications

The process, known as distillation, allows AI developers to learn from a more capable model without building all its capabilities from scratch1

1

. Developers send large numbers of prompts to a stronger model, collect its answers, and use those responses to train their own systems. Done at scale, this technique can help a smaller model replicate advanced capabilities at a fraction of the cost. OpenAI, Anthropic, and Google have argued that distillation enables rivals to reproduce advanced capabilities more quickly and at lower cost without making comparable investments in training2

2

. The practice has become a major flashpoint as competition between US and Chinese AI developers intensifies3

3

.

Previous Accusations Against Moonshot AI

This isn't the first time OpenAI has accused Moonshot AI of using similar methods. On September 30, OpenAI said people linked to the Kimi developer had run a coordinated campaign in July to extract the hidden reasoning behind its models' answers1

1

. OpenAI detected related activity involving more than 15,000 users and shut down the campaign by July 28, banning the accounts involved and sharing its findings with other AI labs1

1

. The latest report adds to geopolitical tensions over how Chinese AI labs develop models that can compete with those built by US companies.

Leading Labs Implement Safeguards Against Distillation Attacks

AI companies are actively working to make distillation harder for competitors. Earlier this month, Anthropic announced that new safeguards had reduced the amount of data collected from its Claude models in distillation attacks by approximately 90% to 99%1

1

2

2

. This development underscores how leading labs are tightening access to their models to prevent unauthorized copying. US agencies have also identified Chinese firms they say targeted American AI models1

1

.

Performance Claims and the Dilemma for AI Leaders

Both Chinese models have posted strong results in their developers' own evaluations. Moonshot AI claimed that Kimi K3.1 Pro outperformed GPT-6 Astra on a coding benchmark1

1

. DeepSeek introduced V4.1-Flash in September as a lower-cost alternative to US models, though neither company's results have been independently verified1

1

. The reported use of Astra highlights a difficult problem for leading AI labs: the same models they offer to users can also serve as training material for competitors looking to catch up1

1

. OpenAI, DeepSeek, and Moonshot AI did not immediately respond to requests for comment2

2

3

3

. The allegations arrive at a sensitive moment for China's AI sector, as DeepSeek and Moonshot AI are among the country's most prominent model developers, and their latest releases have been presented as challengers to leading Western AI models3

3

.