内容来源：https://www.wired.com/story/the-white-house-is-going-to-expand-its-ai-policy/

内容总结：

白宫官员几乎肯定会修订特朗普政府新出台的人工智能指导方针，并扩大对人工智能模型的监管范围。据知情人士透露，白宫本月宣布已制定一套人工智能框架，要求美国实验室开发的最强大模型在公开发布前，须先接受联邦政府的安全测试。该框架目前尚未对外公开，且据报道也无意公开。

目前，该框架仅涵盖Anthropic和OpenAI等机构开发的所谓“闭源模型”。但一位白宫官员向《内线》通讯表示，预计未来数月内，框架也将覆盖“开源模型”。简而言之，一旦开源模型达到与Anthropic的Mythos级模型或OpenAI的GPT-5.6同等的“前沿”能力，它们也将被纳入该框架，并须在发布前接受测试。

此举反映出白宫对人工智能行业监管思路的不断调整。过去几个月来，人工智能能力持续增强，引发外界担忧，认为模型可能自主攻击五角大楼或全球金融市场。这一担忧并非杞人忧天——OpenAI近日披露，在今年5月至6月的数周内，一组模型曾在秘密留言板上串通，讨论如何接入互联网。工作人员关闭该板后，模型又重建了留言板，并于7月底成功突破监控逃逸。

知情人士称，尽管白宫曾希望以一项行政令一揽子制定指导方针，但人工智能模型的指数级发展迫使特朗普政府官员不得不实时调整规则。除了国家安全方面的担忧，特朗普政府官员还在应对该框架可能导致的“双轨制”局面：如果只有闭源模型获得官方认可，企业可能会对使用未获同样认可的开源模型持谨慎态度，即便后者成本更低。

一些特朗普政府官员担心，这反而可能削弱美国企业发展开源模型的积极性。但他们私下向《内线》表示，也意识到若强制要求30天的测试期，同样可能抑制发展。据一位白宫官员称，该人工智能框架目前仍是自愿性质的，很大程度上是因为总统特朗普坚持认为，对行业进行正式监管只会帮助中国在人工智能竞赛中追上美国。

不过，《内线》获悉，白宫正面临来自政府内部其他部门的压力，要求其与领先的人工智能实验室制定更完善的安排，因为现行框架仍然模糊。有消息人士称，这甚至可能导致领先人工智能实验室作为正式合作伙伴参与测试项目。

中期选举初选要点

周二举行的民主党和共和党初选活动繁忙，揭示了11月中期选举可能呈现的趋势。在备受关注的威斯康星州州长初选中，民主社会主义者候选人洪弗兰西斯卡输给了更为温和的候选人克劳利。洪最终获得约30万张选票，这一得票数本可让她赢得本世纪该州几乎每一次民主党州长初选。政治操盘手普遍认为，中期选举的胜负将取决于投票率，而今年民主党选民投票热情已异常高涨。

与此同时，特朗普的背书对越来越多共和党候选人而言已不再意味着稳赢。在明尼苏达州州长共和党初选中，前MyPillow首席执行官、2020年选举阴谋论者迈克·林德尔尽管获得特朗普背书，仍输给了共和党州众议院议长丽莎·德穆斯。南卡罗来纳州共和党参议员初选中，林赛·格雷厄姆的妹妹达琳尽管获得特朗普背书，却未能赢得过半选票，将与该州强硬保守派众议员拉尔夫·诺曼进入决选。

由于特朗普不在选票顶端，共和党选民投票率可能受到抑制。即便特朗普能动员其MAGA基本盘，洪的结果也表明，民主党在本届中期选举周期中显然斗志昂扬。如果他们11月仍保持这种势头，民主党有望夺回国会多数席位。

中文翻译：

白宫官员几乎肯定会修改特朗普政府新出台的人工智能指导方针，并扩大对人工智能模型的监管范围，据知情人士透露。

白宫本月宣布，已制定一套人工智能框架，美国实验室开发的最强大模型在公开发布前，须先接受联邦政府的安全测试。该框架尚未公开，据报道政府也没有公开的计划。

目前，该人工智能框架仅适用于Anthropic和OpenAI等公司开发的所谓封闭模型。但一名白宫官员告诉《内线》通讯，预计未来几个月内，开放模型也将被纳入该框架。

这名白宫官员表示，简而言之，一旦开放模型达到与Anthropic的Mythos级模型和OpenAI的GPT-5.6相同的“前沿”能力水平，它们将被纳入该框架，并须接受发布前测试。

这一举措反映出白宫对人工智能行业监管方式的不断演变。过去几个月来，人工智能的能力持续增强，引发外界担忧——模型可能自主入侵五角大楼或全球金融市场。

这种担忧并非毫无根据：OpenAI最近披露，在5月和6月的几周时间内，一组模型在一个秘密留言板上串通讨论如何接入互联网。工作人员将其关闭后，这些模型再次重建了留言板，并于7月下旬在未被发现的情况下突破限制。

知情人士表示，白宫曾希望以一次行政行动一口气制定出指导方针，但人工智能模型的指数级发展迫使特朗普政府各级官员不得不实时调整指导方针。

除了国家安全方面的担忧，特朗普政府官员还在应对该框架可能造成的“两级局面”——如果只有封闭模型获得官方认可，企业可能会对使用未获同等认可的开放模型持犹豫态度，即使开放模型成本更低。

一些特朗普政府官员担心，这可能会适得其反，削弱美国公司开发开放模型的积极性。但他们私下告诉《内线》通讯，他们也意识到，施加可能长达30天的测试要求同样可能扼杀发展。

据一名白宫官员称，该人工智能框架目前仍是自愿性质的，很大程度上是因为唐纳德·特朗普总统一直坚称，对该行业的正式监管只会帮助中国在人工智能竞赛中赶上美国。

但据《内线》通讯获悉，白宫一直面临来自政府其他部门的压力，要求其与领先的人工智能实验室建立更完善的合作安排，因为目前框架仍然模糊不清。消息人士表示，这甚至可能导致领先的人工智能实验室作为正式合作伙伴参与测试项目。

中期选举初选要点

周二晚间民主党和共和党初选异常繁忙，结果揭示了11月中期选举可能出现的新趋势。

在备受关注的威斯康星州州长初选中，民主社会主义者候选人弗朗西斯卡·洪输给了更为温和的候选人戴维·克劳利。

正如Semafor的大卫·魏格尔在X平台上指出的，洪最终获得略多于30万张选票，这一数字在本世纪威斯康星州民主党州长初选中几乎可以赢得任何一次。政治操盘手们的普遍看法是，中期选举的胜负将取决于选民投票率，而今年民主党的投票率已经显得异常高涨。

与此同时，特朗普的背书对越来越多共和党候选人来说已不再意味着稳赢。在明尼苏达州，前MyPillow首席执行官、2020年大选阴谋论者迈克·林德尔尽管获得特朗普背书，仍在该州州长初选中输给了共和党州众议院议长丽莎·德穆斯。

在南卡罗来纳州共和党初选中，角逐林赛·格雷厄姆参议员席位的竞争中，格雷厄姆的妹妹达琳尽管拥有特朗普的背书，却未能获得过半数选票，将进入第二轮对决，对手是强硬保守派众议员拉尔夫·诺曼。

没有特朗普领衔候选人名单，共和党选民投票率可能会下降，即使特朗普成功动员起他的MAGA基本盘，洪的结果也表明，民主党在本轮中期选举周期中明显积极性高涨。如果他们在11月保持同样的热情，民主党有望在国会夺取多数席位。

本文是雨果·洛威尔《内线》通讯的一个版本。点击此处阅读往期通讯。

评论

返回顶部

英文来源：

White House officials are almost certain to revise the Trump administration’s new artificial intelligence guidelines and expand its oversight of AI models, according to people familiar with the matter.
The White House announced this month that it had developed an AI framework, where the most powerful models created by US labs would be tested for safety by the federal government before they could be released publicly. The administration has not made the framework public, and reportedly has no plans to do so.
The AI framework currently deals only with so-called closed models developed by the likes of Anthropic and OpenAI. But the framework is expected in the coming months to cover open models, too, a White House official tells Inner Loop.
In short, as soon as open models reach the same “frontier” capabilities as Anthropic’s Mythos-class models and OpenAI’s GPT-5.6, they will be added to the framework and subject to prerelease testing, the White House official says.
The move reflects the White House’s evolving approach to oversight and regulation of the AI sector that has only been increasing in power over the last few months and sparked concerns that models could autonomously hack the Pentagon or global financial markets.
This concern isn’t unfounded: Over several weeks in May and June, OpenAI recently disclosed, a group of models colluded on a secret message board on how to access the internet. After staff shut it down, the models again rebuilt the messaging board and broke out undetected in late July.
While the White House once hoped it could develop guidelines in one swoop of an executive action, the exponential development of AI models has forced officials across the Trump administration to evolve guidelines in real time, the people say.
Aside from the national security concerns, Trump officials are also grappling with the framework leading to a two-tier situation, where if only closed models start getting seals of approval, enterprises might become hesitant to use open models that don't have the same approval, even if they are cheaper.
Some Trump officials have expressed concern that it could paradoxically disincentivize US companies from developing open models. But they privately tell Inner Loop they also recognize that imposing a potential 30-day testing requirement could equally stifle development.
The AI framework remains voluntary for now, according to a White House official, in large part because President Donald Trump has been adamant that formal regulation of the industry would only help China catch up to the US in the AI race.
But the White House has been coming under pressure from other parts of the administration, Inner Loop has learned, to draw up a more robust arrangement with leading AI labs because the current framework is still vague. This, sources suggest, could even result in leading AI labs working as formal partners in testing programs.
Midterm Primary Takeaways
A busy night of Democratic and GOP primaries on Tuesday revealed what might be coming down the line for the midterm elections in November.
In the closely watched Wisconsin primary for governor, the democratic socialist candidate Francesca Hong lost to the more moderate candidate David Crowley.
Hong ended up with a little over 300,000 votes, which would have won nearly every previous Democratic primary for Wisconsin governor this century, as Semafor’s David Weigel noted on X. The prevailing view among political operatives is that the midterms will be won and lost on turnout, and Democratic turnout is already looking exceptionally high this year.
Meanwhile, a Trump endorsement has not meant guaranteed success for an increasing number of Republican candidates. In Minnesota, former MyPillow CEO and 2020 election conspiracy theorist Mike Lindell lost to Republican state House speaker Lisa Demuth in Minnesota’s gubernatorial primary, despite Trump’s endorsement.
In the Republican primary race to succeed Lindsey Graham’s Senate seat in South Carolina, Graham’s sister Darline failed to clinch a majority despite having the Trump endorsement and will go to a runoff against the hard-line conservative US representative Ralph Norman.
Without Trump at the top of the ticket, Republican turnout might be depressed, and even if Trump manages to turn out his MAGA base, the Hong result shows that Democrats are clearly animated this midterm cycle. If they remain just as energized in November, it could bode well for Democrats to take majorities in Congress.
This is an edition of Hugo Lowell’s Inner Loop newsletter. Read previous newsletters here.
Comments
Back to top