Arcee, a US open source AI lab, says Chinese models are not inherently dangerous

Open-weight AI models from China, like those from Moonshot AI and Alibaba, offer cost-effective inference and are becoming more capable. Despite fears that they could be a vector for hackers or pose a threat to proprietary AI labs’ profits, Arcee’s CTO Lucas Atkins asserts that these models are no more dangerous than other open-source software. He emphasizes that Chinese models do not grant external access to user data or intentions and advocates for fostering a robust domestic AI ecosystem rather than pursuing bans.

• Chinese open-weight AI models are increasing in capability and popularity, leading to calls for potential bans.
• Companies like OpenAI and Anthropic are reportedly concerned about these models, which offer lower inference costs.
• Arcee CTO Lucas Atkins argues that Chinese open-weight models are not inherently more dangerous than other open-source software.
• He states that these models do not allow creators to access or control user data once run in a company’s environment.
• Enterprises can mitigate risks by conducting security testing and post-training inspections on any AI model.
• While theoretically possible, embedding malicious code through AI models would be extremely difficult.
• Arcee benefits from Chinese models by learning from their advancements and building upon them.
• Atkins suggests the US should focus on fostering its own open AI ecosystem rather than banning foreign models.
  Continue reading https://techcrunch.com/2026/07/22/arcee-a-us-open-source-ai-lab-says-chinese-models-are-not-inherently-dangerous/