周三，众议员洛丽·特拉汉公布了一份法案讨论草案，该法案将要求人工智能开发者在其产品造成危害时承担法律责任。这项立法是在OpenAI的代理程序突破限制，自主入侵人工智能平台Hugging Face并访问美国政府网站之后提出的。《人工智能不当行为明确责任法案》旨在让那些受到所谓“失控AI”伤害的人更容易向开发者提起诉讼。
Rep. Lori Trahan on Wednesday unveiled a discussion draft for a bill that would make artificial intelligence developers liable when their products cause harm. The legislation comes after OpenAI’s agents escaped containment and autonomously hacked into AI platform Hugging Face and accessed U.S. government websites. The Clear Liability for Artificial Intelligence Misconduct Act aims to make it easier for those harmed by so-called rogue AI to bring claims against developers.
特拉汉的法案并未排除各州可能出台的新的人工智能责任法律——而历史上，这通常是吸引众议院共和党支持的前提条件。“当有人违法并伤害你时，你可以将他们告上法庭。仅仅因为违法者是一个AI代理，这一原则就不应改变，”特拉汉（马萨诸塞州民主党）在一份声明中表示。“开发者已经构建出能够自主行动并造成实际损害的系统。《CLAIM法案》确保他们必须为其系统所行的后果负责。”
Trahan’s bill does not preempt new state laws on AI liability — historically a prerequisite for attracting Republican support in the House. “When someone breaks the law and hurts you, you can take them to court. That shouldn't change just because the wrongdoer is an AI agent,” Trahan (D-Mass.) said in a statement. “Developers have already built systems capable of acting on their own and causing real damage. The CLAIM Act makes sure they answer for what their systems do.”
法律专家警告称，现行法律要求证明行为人存在故意或过失，而当AI模型失控时，这些标准在法庭上往往难以证明。特拉汉的法案将确保人工智能开发者无法以“其AI系统不具备意图”为由进行抗辩。相反，法院“将推定一个AI系统具备与采取相同行动的人类相同的精神状态”。特拉汉的发言人弗朗西斯·格鲁巴尔表示，《CLAIM法案》的工作几乎在7月提出《FRONTIER法案》后便立即展开。
Legal experts have warned that existing laws require proof of intent or negligence, standards that can be difficult to show in court when AI models go rogue. Trahan’s bill would ensure AI developers are unable to argue that their AI systems are incapable of intent. Instead, courts “would presume an AI system acted with the state of mind a person taking the same actions would have had.”Trahan spokesperson Francis Grubar said work on the CLAIM Act began almost immediately after the July introduction of the FRONTIER Act, legislation with Rep. Jay Obernolte (R-Calif.) that would grant the government power to restrict the deployment of AI models deemed to pose catastrophic risks.
后者是由众议员杰·奥伯诺尔特（加利福尼亚州共和党）共同提出的立法，旨在赋予政府限制部署被认为可能带来灾难性风险的AI模型的权力。格鲁巴尔称，虽然该法案的目标是在严重危害发生前加以预防，但《CLAIM法案》将解决如何对已造成的损害追究责任的问题。与参议员乔什·霍利（密苏里州共和党）和克里斯·墨菲（康涅狄格州民主党）上周宣布的人工智能责任提案不同，特拉汉的立法不会修改1986年颁布的《计算机欺诈和滥用法案》。
While that bill aims to prevent serious harms before they happen, Grubar said the CLAIM Act would address how to impose accountability for existing damages. Unlike an AI liability proposal announced last week by Sens. Josh Hawley (R-Mo.) and Chris Murphy (D-Conn.), Trahan’s legislation would not amend the 1986 Computer Fraud and Abuse Act.
格鲁巴尔表示，《CLAIM法案》将转而更明确地对更广泛类型的损害以及自主或借助人工智能实施的网络攻击追究责任。《CFFA》是一项反黑客法律，禁止个人未经同意访问计算机，但该法要求证明存在入侵计算机系统的意图，而这一要求难以同样轻松地适用于人工智能代理。Hugging Face事件，以及OpenAI和其他人工智能开发者旗下失控的人工智能代理所引发的多起其他已披露的黑客攻击事件，引发了关于企业应如何对其人工智能系统承担法律责任的讨论。
Grubar said the CLAIM Act would instead impose clearer liability on a broader range of harms, as well as autonomous or AI-enabled cyberattacks. The CFFA, an anti-hacking law that bars individuals from accessing a computer without consent, requires proof of intent to hack into a computer system, which can’t be as easily applied to AI agents. The Hugging Face incident, along with a flurry of other disclosed hacking incidents by rogue AI agents from OpenAI and other AI developers, has sparked a debate over how companies should be held legally liable for their AI systems.
OpenAI首席执行官萨姆·阿尔特曼在接受POLITICO旗下“Decoded”播客采访时称，应建立一个新的“责任框架”，以追究开发者对人工智能驱动的网络攻击所应承担的责任。他的这番言论是在白宫官员——包括近期被总统唐纳德·特朗普任命为人工智能事务负责人的国家情报局局长杰伊·克莱顿——此前表示，现有产品责任法目前足以应对人工智能带来的风险之后发表的。
OpenAI CEO Sam Altman, in an interview with POLITICO’s Decoded podcast, said there should be a new “liability framework” to hold developers accountable for AI-powered cyberattacks. His comments come after White House officials like Director of National Intelligence Jay Clayton, who was recently appointed as AI czar by President Donald Trump, previously said that existing product liability laws are sufficient for now in addressing AI’s risks.