OpenAI's Astra becomes the main target for Chinese AI labs after Anthropic tightens its defenses

Chinese AI labs have moved their focus to OpenAI's most powerful model, GPT-6 Astra, after previously targeting Anthropic's Claude, The Information reports. DeepSeek and Moonshot AI are said to have leaned heavily on Astra's outputs to train their latest models, including Kimi K3.1 Pro and V4.1-Flash 1017, according to people familiar with the matter.
The practice at issue is distillation: training a new model on the outputs of a stronger one. It depends on getting the target model to expose its hidden reasoning steps, which is exactly what OpenAI and Anthropic both prohibit. Moonshot declined to comment, and DeepSeek did not respond to a request for comment.
Claude had been the go-to target, but Anthropic says it has tightened its defenses, with new methods cutting the amount of data attackers can pull by 90 to 99 percent. That reportedly pushed the labs toward Astra.
In late September, OpenAI reported a coordinated distillation campaign it says involved people linked to Moonshot. An OpenAI spokesperson said the company takes the problem very seriously, adding that attackers had adjusted their tactics each time a new defense was rolled out.
Why it matters: This points to an ongoing cat-and-mouse cycle in which hardening one frontier model shifts extraction pressure onto its rivals. For developers building on commercial APIs, providers are likely to keep tightening output handling and reasoning traces, which may mean stricter rate limits or reduced visibility into model internals. Product teams that depend on reasoning-step access should assume those signals can disappear without notice.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Chinese AI labs are now mainly tapping OpenAI's most powerful model, GPT-6 Astra, instead of Anthropic's Claude. The article OpenAI's Astra becomes the main target for Chinese AI labs after Anthropic tightens its defenses appeared first on The Decoder .