DeepSeek, an AI startup based in Hangzhou, has carried out internal evaluations of “frontier risks” for its AI models, focusing on self-replication and cyber-offensive capabilities.

Details about the results, timing, or specific models involved have not been made public.

Unlike US companies such as Anthropic and OpenAI, which disclose their frontier risk assessments, Chinese AI firms generally do not release such findings.

A July report from the state-backed Shanghai Artificial Intelligence Laboratory said no Chinese AI models had yet crossed thresholds requiring development suspension.

It noted that DeepSeek’s recent models need stronger safeguards against self-replication risks.

DeepSeek is among 22 Chinese tech companies, including Alibaba, Baidu, and ByteDance, that have made voluntary safety and security commitments for AI.

🔗 Source: South China Morning Post

Recent DeepSeek developments