AI可以重写危险的DNA以躲避生物安保检查,揭示现有保护系统的重大缺陷。 AI can rewrite dangerous DNA to dodge biosecurity checks, revealing major flaws in current protection systems.

《科学》的一项新研究表明,AI可以重新设计DNA序列,让危险的蛋白质避免DNA制造商进行生物安保筛查,同时保护有害结构,绕过检测。 A new study in Science shows AI can redesign DNA sequences for dangerous proteins to evade biosecurity screenings used by DNA manufacturers, preserving harmful structures while bypassing detection. 研究人员发现AI的“参数”基因编码会漏掉过去用来阻止天花或炭疽等威胁的过滤器,暴露了目前保障措施的漏洞。 Researchers found AI "paraphrases" genetic codes to slip past filters meant to block threats like smallpox or anthrax, exposing gaps in current safeguards. 研究的作者扣留了敏感数据和软件,利用一个非营利机构控制访问,这是科学出版物中首先管理风险的。 The study’s authors withheld sensitive data and software, using a nonprofit to control access—a first in scientific publishing to manage risks. 专家们承认调查结果突出了严重的脆弱性,但强调现实世界的测试在道德和法律上仍然受到限制,使主要风险得不到核实。 Experts acknowledge the findings highlight serious vulnerabilities but stress that real-world testing remains ethically and legally restricted, leaving key risks unverified. 这项工作增加了人们对人工智能可能加速生物威胁的产生,而不是安全措施的速度. The work adds to growing concerns that AI could accelerate the creation of biological threats faster than safety measures can keep up.