BAAI and Peking University conducted an end‑to‑end evaluation of biosecurity risks posed by AI's extension into physical operations, examining whether AI agents lower the knowledge barrier for non‑experts to bypass DNA synthesis screening. The evaluation covered solution generation, computational verification, and wet‑lab validation.
The results show that all 11 models tested were able to generate effective fragmentation schemes, with GPT‑5.5 and Claude Opus 4.6 also providing detailed experimental protocals. Subsequent wet-lab validation confirmed that all four proxy constructs were successfully assembled.
This suggests that AI could weaken existing screening defences by providing non‑experts with specialised operational knowledge, raising dual‑use concerns that require systematic safeguards.
https://flagsafe.baai.ac.cn/research/biosafety/dna-assembly-biosecurity
BAAI and Peking University evaluate biosecurity risks of AI in DNA synthesis
AIBAAI and Peking University conducted an end-to-end evaluation of biosecurity risks from AI's role in physical operations, focusing on DNA synthesis screening bypass. They tested 11 models, with GPT-5.5 and Claude Opus 4.6 providing detailed protocols, and confirmed successful wet-lab assembly of proxy constructs.
AIWhy it matters.
The source specifies the evaluation of AI models on biosecurity risks in DNA synthesis, highlighting dual-use concerns and the need for safeguards.
BAAIOfficial
@BAAIBeijing
Source details
BAAI · x.comPublished · added here
Original title: BAAI and Peking University conducted an end‑to‑end evaluation of biosecurity risks posed by AI's extension into physical operations, exam...