随著人工智慧在科学领域的应用日益广泛，Anthropic、OpenAI 与 Google DeepMind 等领先企业正加紧防范未来模型被用于制造新型病毒或生物武器的风险。由于化学与生物技术具备高度军民双用性，用于开发疫苗的知识同样可能被滥用于制造生物武器，业界普遍认为生物安全将成为继网路安全后的下一个重大公共安全威胁。

过去如奥姆真理教发动东京地铁沙林毒气攻击等历史事件，以及近期波哥大圣战组织（Boko Haram）利用生成式 AI 工具辅助武器工程的研究，皆引发各界对 AI 降低恐怖攻击与致命病原体设计门槛的忧虑。为降低潜在危害，AI 厂商除明文禁止相关用途外，亦透过可信存取计划审核使用者权限、部署自动防护机制以拦截可疑查询，并与 SecureBio 等外部评估机构合作进行安全测试。（关键数字：1995）

然而，专家指出生物风险评估面临极大挑战，因验证模型危害往往涉及高风险且昂贵的湿实验室测试，无法像网路安全般在数位环境中进行端对端模拟。目前产学界对 AI 在生物领域的具体风险程度仍未达成共识，但学者警告大众与决策者尚未做好充分准备，任何单一重大事故都可能重创公众对科学与医疗技术的信任，因此在追求技术效益的同时必须保持审慎。

As artificial intelligence becomes increasingly capable in scientific domains, leading AI groups including Anthropic, OpenAI, and Google DeepMind are intensifying efforts to mitigate the risks of future models aiding in the creation of novel viruses or biological weapons. Because chemical and biological data is inherently dual-use—where information used for vaccine development can also be repurposed for bioweapons—experts view biosecurity as the next major public safety frontier after cybersecurity.

Historical precedents, such as the 1995 Tokyo subway sarin attack by Aum Shinrikyo, alongside recent findings showing Boko Haram militants using generative AI for weapons engineering, have heightened alarms that advanced models could significantly lower technical barriers for bad actors to deploy harmful pathogens. In response, AI developers prohibit terrorist applications, enforce trusted access vetting for advanced features, deploy query-filtering guardrails, and collaborate with external evaluation groups like SecureBio to test model capabilities.

Nevertheless, experts highlight that evaluating biological risks remains exceedingly difficult, as direct empirical verification often requires slow, costly, and hazardous wet-lab experiments rather than purely digital simulations. While a scientific consensus on the precise severity of AI biosecurity risks has yet to form, researchers warn that society and policymakers remain largely unprepared, emphasizing that a single catastrophic failure could permanently damage public trust in scientific and technological advancements unless rigorous prudence is maintained.