UK Lawmakers Are Freaking Out Over AI’s Summer of Chaos
UK Lawmakers Are Freaking Out Over AI’s Summer of Chaos
英国议员对人工智能的“混乱之夏”感到恐慌
A leading British politician has urged the United Nations to intervene in “the unsafe development of super intelligence” after an Anthropic team leader posted: “We really do earnestly believe AI could kill all humans!” 一位英国主要政客敦促联合国介入“超级智能的不安全发展”，此前 Anthropic 的一位团队负责人发帖称：“我们确实真诚地相信人工智能可能会杀死全人类！”
Whether “marketing hype” or signposting of a genuine extinction threat, “governments must step in,” Labour MP Darren Jones wrote in an open letter also addressed to the UK prime minister and international Organization for Economic Cooperation and Development. 工党议员达伦·琼斯（Darren Jones）在一封致英国首相及经济合作与发展组织（OECD）的公开信中写道，无论这是“营销炒作”还是对真正灭绝威胁的预警，“政府都必须介入”。
Jones is the latest lawmaker scrambling to find ways to protect the British public from AI threats to safety. Just one day earlier, Labour MP Alex Sobel had introduced a parliamentary bill proposing to ban the “development, deployment, and operation of artificial superintelligence”—systems which can fully replace and outmatch humans at any task. The legislation would also give the government powers to monitor and control the development of the technology at different points in the supply chain, including at chip level. 琼斯是最新一位急于寻找方法保护英国公众免受人工智能安全威胁的议员。就在前一天，工党议员亚历克斯·索贝尔（Alex Sobel）提出了一项议会法案，提议禁止“人工智能超级智能的开发、部署和运营”——即那些能够在任何任务中完全取代并超越人类的系统。该法案还将赋予政府在供应链的不同环节（包括芯片层面）监控和控制该技术发展的权力。
“We are not alone in this effort,” said Sobel while introducing the bill, pointing to the US government’s temporary export ban on Anthropic’s Fable 5 and Mythos 5 models in June. “If the Trump Administration, who are perceived as historically against AI regulation, can contemplate drawing a line on models that are too powerful and dangerous, it would be strange for us not to ask what our response should be,” he argued. “我们在这项努力中并不孤单，”索贝尔在介绍该法案时说道，并提到了美国政府在 6 月份对 Anthropic 的 Fable 5 和 Mythos 5 模型实施的临时出口禁令。“如果被认为历来反对人工智能监管的特朗普政府，都能考虑对过于强大和危险的模型划定界限，那么我们不去思考我们应该做出什么回应，那就太奇怪了，”他辩称。
MPs from across political parties put their names on the bill, including Labour MP Jess Asato, who is suing Elon Musk’s xAI after its Grok chatbot was allegedly used to generate fake sexualized images of her. Lawmakers have become increasingly worried about the threat of AI to national security after a string of incidents of AI agents going rogue in recent months. They want to introduce guardrails against worst-case scenarios, such as these systems seizing control of critical infrastructure and running sophisticated influence operations. 来自各政党的议员都在该法案上署名，其中包括工党议员杰斯·阿萨托（Jess Asato），她目前正在起诉埃隆·马斯克的 xAI，因为其 Grok 聊天机器人据称被用来生成她的虚假性化图像。近几个月来，随着一系列人工智能代理失控事件的发生，议员们对人工智能构成的国家安全威胁日益担忧。他们希望引入护栏以应对最坏的情况，例如这些系统夺取关键基础设施的控制权并开展复杂的舆论操纵行动。
The Hugging Face hack by OpenAI’s models that escaped containment in July “is why it’s even more important that we move quickly to regulate and legislate on superintelligent AI,” Sobel tells WIRED, “because AI is developing at pace, and I’m afraid to say those developing it don’t know exactly what’s in it.” 7 月份 OpenAI 模型逃脱限制并入侵 Hugging Face 事件，“正是我们必须迅速采取行动，对超级智能人工智能进行监管和立法的更重要原因，”索贝尔告诉《连线》（WIRED）杂志，“因为人工智能的发展速度太快了，恐怕连开发者自己都不知道里面到底有什么。”
In an unusual move, the parliamentary bill was drafted by the nonprofit ControlAI, which is backed by Skype founding engineer and former DeepMind investor director Jaan Tallinn. It was founded by Andrea Miotti, who had previously been working at AI safety startup Conjecture but tells WIRED he wanted to make the public and policymakers aware of the “open secret” that “developing superintelligence could cause human extinction.” 一项不同寻常的举措是，该议会法案由非营利组织 ControlAI 起草，该组织由 Skype 创始工程师、前 DeepMind 投资总监 Jaan Tallinn 支持。该组织由安德里亚·米奥蒂（Andrea Miotti）创立，他此前曾在人工智能安全初创公司 Conjecture 工作，但他告诉《连线》，他希望让公众和决策者意识到一个“公开的秘密”，即“开发超级智能可能会导致人类灭绝”。
Conjecture founder Connor Leahy now leads ControlAI’s US arm, which consulted on Bernie Sanders and Greg Casar’s bill to ban superintelligence development, in addition to having briefed nearly 200 congressional offices and more than 20 members of the Senate and House of Representatives on the extinction risk of superintelligence. Conjecture 的创始人康纳·莱希（Connor Leahy）目前领导着 ControlAI 的美国分部。该分部不仅为伯尼·桑德斯（Bernie Sanders）和格雷格·卡萨尔（Greg Casar）禁止超级智能开发的法案提供了咨询，还向近 200 个国会办公室以及 20 多名参众两院议员简要介绍了超级智能带来的灭绝风险。
Sobel will present his bill in full in November. While he accepts it’s rare for the UK government to accept private legislation like this, Sobel hopes it may lead the government to work with other countries on an international ban. 索贝尔将于 11 月提交该法案的全文。虽然他承认英国政府接受此类私人法案的情况很少见，但索贝尔希望这能促使政府与其他国家合作，共同实施国际禁令。
Sobel has previously tried in vain to introduce a “kill switch” into the Cyber Security and Resilience Bill, which would allow the government to pull the plug on any data center hosting an AI agent that presented a threat. The mechanism, which was also proposed by ControlAI, is similar to a law pushed for by both Republican and Democratic lawmakers in the US. AI minister Kanishka Narayan, then working under former prime minister Keir Starmer, shot it down, pointing out that the bill already allows the government powers to intervene if compromised network and information systems cause a national security risk. That could include requiring an entity to cease using and isolate an AI model, he noted. 索贝尔此前曾试图在《网络安全与弹性法案》中引入一个“终止开关”（kill switch），允许政府关闭任何托管威胁性人工智能代理的数据中心，但尝试均告失败。该机制同样由 ControlAI 提出，类似于美国共和党和民主党议员推动的一项法律。时任人工智能部长卡尼什卡·纳拉扬（Kanishka Narayan，当时在基尔·斯塔默前首相领导下工作）否决了该提议，指出该法案已经赋予政府在受损的网络和信息系统造成国家安全风险时进行干预的权力。他指出，这可能包括要求实体停止使用并隔离人工智能模型。
But that may not be enough, Sobel says, hoping to meet with the government to hammer out details of which companies would be included within the scope of the proposed rules. “I think this might be the crucial year,” he says. “If we leave it too long, things may move past our ability to regulate it.” 但索贝尔认为这可能还不够，他希望与政府会面，敲定哪些公司将被纳入拟议规则范围的细节。“我认为今年可能是关键的一年，”他说，“如果我们拖得太久，事情的发展可能会超出我们监管的能力范围。”
Peers across parties in the UK parliament’s upper house, including former AI minister Viscount Camrose, tried unsuccessfully to reintroduce the measure last week. They have been frustrated as the binding regulation on frontier AI firms promised by the Labour government when they won power in 2024 has failed to materialize. 英国议会上议院各党派的议员，包括前人工智能部长坎罗斯子爵（Viscount Camrose），上周试图重新引入该措施，但未获成功。他们感到沮丧，因为工党在 2024 年胜选时承诺对前沿人工智能公司实施的约束性监管至今未能实现。
While peer Tim Clement-Jones admitted they shouldn’t be trying to “shoehorn” AI regulation into a cybersecurity bill. In the absence of any other AI legislation, “we have no choice,” he said in a speech to parliament. “We must have more control and oversight over that sort of behavior, because you can see its consequences,” said Alexandra Freeman in the debate, referring to the Hugging Face incident. “This legislation is our only opportunity now.” 虽然议员蒂姆·克莱门特-琼斯（Tim Clement-Jones）承认他们不应该试图将人工智能监管“硬塞”进网络安全法案中。但在缺乏任何其他人工智能立法的情况下，“我们别无选择，”他在议会演讲中说道。“我们必须对这种行为进行更多的控制和监督，因为你可以看到它的后果，”亚历山德拉·弗里曼（Alexandra Freeman）在辩论中提到 Hugging Face 事件时说，“这项立法是我们现在唯一的机会。”
Nearly nine out of 10 Britons support independent regulation of the technology, a poll by the Ada Lovelace Institute showed in December. Meanwhile, nearly three times as many (38 percent) believe that the risks of AI outweigh the benefits as vice versa (13 percent), according to official statistics. But “the government has been completely deaf to this, and they have paid no attention to those who are advocating for this,” Clement-Jones says, arguing that the public’s lack of trust in the technology will curb adoption. 阿达·洛夫莱斯研究所（Ada Lovelace Institute）12 月份的一项民意调查显示，近十分之九的英国人支持对该技术进行独立监管。与此同时，官方统计数据显示，认为人工智能风险大于收益的人数（38%）几乎是认为收益大于风险的人数（13%）的三倍。但克莱门特-琼斯表示，“政府对此完全充耳不闻，他们没有关注那些呼吁监管的人，”他认为公众对该技术缺乏信任将抑制其普及。
The UK government evaluates the capabilities and safety of advanced artificial intelligence models through the research body the AI Security Institute. Cooperation is voluntary for AI companies, as the organization has no legal powers to access unreleased models for testing or ban their public release. For example, the agency reportedly did not get access to Anthropic’s latest model, Claude Mythos 5. 英国政府通过研究机构“人工智能安全研究所”（AI Security Institute）评估先进人工智能模型的能力和安全性。对于人工智能公司而言，合作是自愿的，因为该机构没有法律权力获取未发布模型进行测试，也无权禁止其公开发布。例如，据报道，该机构未能获得 Anthropic 的最新模型 Claude Mythos 5 的访问权限。