内容来源：https://www.wired.com/story/uncanny-valley-podcast-ai-ipo-race-elon-musk-doge-whistleblower-instagram-hacking-incident/

内容总结：

AI公司IPO竞赛白热化，旧金山楼市惊现“股票换房产”

本周，一场围绕顶级AI公司IPO的竞赛正在硅谷掀起波澜。Anthropic已秘密提交上市文件，估值高达9650亿美元，有望成为史上规模最大的IPO之一。紧随其后，SpaceX（旗下拥有xAI）也已提交上市申请，OpenAI据传也将很快跟进。

这场IPO热潮已产生奇特连锁反应：旧金山多套房产挂牌时明确表示，愿意接受Anthropic或OpenAI的股票代替现金支付。有房产经纪透露，许多Anthropic员工对此表现出浓厚兴趣，甚至有卖家因更青睐Anthropic的产品而只接受其股票。

特朗普签署AI行政令，效果遭质疑

本周，特朗普签署了一项关于AI安全的新行政令，要求AI公司在发布新模型前30天自愿向政府提供访问权限。与此前的90天方案相比，该令被批评为“力度不足”。据悉，特朗普曾在上个月最后一刻否决了原提案，但最终在白宫高级顾问的劝说下签署了这份折中方案。有分析认为，该行政令本质上是“自愿性质的”，难以称之为真正的监管。

Instagram AI聊天机器人遭黑客利用

近期，多名Instagram用户账号被黑客入侵，其中包括美国前总统奥巴马和白宫使用过的官方账号。令人震惊的是，黑客仅通过向Instagram的AI聊天机器人提出添加邮箱地址的请求，就轻松获得了账号访问权限。网络安全专家警告，随着更多公司将客户服务交给AI代理，这类系统性安全漏洞可能会更频繁地出现。

DOGE吹哨人起诉马斯克诽谤

一名曾在国家劳资关系委员会工作的吹哨人近期对马斯克提起诽谤诉讼。该吹哨人此前指控DOGE团队窃取该机构约10GB敏感数据，马斯克随后在X平台上转发了称其“作伪证”的帖子。该吹哨人声称，在他公开举报后，不仅收到了威胁纸条，其车辆还发现刹车被切断、安全气囊被禁用。他已向华盛顿特区法院提起诉讼，指控马斯克的言论使其生命面临严重威胁。

中文翻译：

本周在《Uncanny Valley》节目中，我们将探讨为何顶尖AI公司的上市竞赛才刚刚开始，却已引发离奇连锁反应——包括旧金山多处房产挂牌时更倾向于接受Anthropic股票而非现金。我们还将解析为何特朗普关于AI安全的新行政令令人失望，以及黑客如何利用Instagram的AI聊天机器人入侵奥巴马总统等知名人士的账号。此外，我们将报道一名DOGE举报人起诉埃隆·马斯克诽谤——在这位亿万富翁公开称其为骗子后，他的生命便受到威胁。

本期提及文章：

• 旧金山房地产中比现金更值钱的是什么？Anthropic股票
• 特朗普终于签署AI行政令的始末
• 他揭发DOGE内幕，随后刹车被剪断

您可在Bluesky上关注Brian Barrett（@brbarrett）、Zoë Schiffer（@zoeschiffer）和Leah Feiger（@leahfeiger）。来信请发送至[email protected]。

收听指南
您可通过本页音频播放器收听本周播客，如需免费订阅每期节目，请按以下步骤操作：
若使用iPhone或iPad，请打开"播客"应用，或直接点击此链接。您也可下载Overcast或Pocket Casts等应用，搜索"uncanny valley"。我们也在Spotify上线了。

文字记录
注：此为自动生成文字稿，可能存在错误。

Brian Barrett：大家好，我是Brian。开始前先说两件事。如果您喜欢我们的节目，请在您的播客应用中花几秒钟为我们评分，这能帮助我们触达更多听众。其次，如果您有关于科技、隐私或政治的任何问题想让我们（我、Zoë和Leah）解答，请现在发送至[email protected]。问题不分大小，我们期待您的来信并为您解惑。好，节目开始。

Zoë Schiffer：大家好。
Brian Barrett：各位好。今天我们分散各地。
Zoë Schiffer：是啊，你本不该出现在这里的，但你还是来了。
Brian Barrett：不能错过。我这周本该休假，和父母家人在海滩度假，打了许多迷你高尔夫。就这样。
Leah Feiger：听起来不错啊。谁赢了？
Brian Barrett：奶奶，奶奶大杀四方。
Zoë Schiffer：哇。
Brian Barrett：球场上的恐怖分子。
Zoë Schiffer：欢迎收听《连线》杂志的《Uncanny Valley》。我是Zoë Schiffer，商业与产业总监。
Brian Barrett：我是Brian Barrett，执行主编。
Leah Feiger：我是Leah Feiger，政治与科学总监。
Zoë Schiffer：今天节目我们将深入探讨顶尖AI公司掀起的IPO热潮——这股热潮甚至让旧金山部分房产挂牌时不再满足于普通现金，而是想要当前更值钱的东西：Anthropic和OpenAI的股票。这场疯狂才刚刚开始。
Leah Feiger：本周特朗普政府签署行政令，要求AI公司在公开发布最先进模型前30天向联邦政府开放访问权限。这一决定并非一帆风顺。我们将讨论政府内部围绕AI监管的内斗，以及这项行政令是否真能产生影响。
Brian Barrett：我们还会聊聊黑客如何欺骗Instagram的AI聊天机器人入侵包括奥巴马前白宫账号在内的高知名度账户。此外，我们将讲述一名DOGE举报人在埃隆·马斯克公开称其为骗子数日后，发现汽车刹车被剪断的遭遇。

Zoë Schiffer：大家可能已听说，本周Anthropic正式提交了上市文件。它以保密方式提交后，便高调对外宣布。
Brian Barrett：而且他们抢在了OpenAI前面。
Zoë Schiffer：对，抢先OpenAI一步，这点我们稍后会细说。该公司尚未确定具体上市日期和募资金额，但以其当前高达9650亿美元的估值——没错，是"亿"为单位——它正稳步迈向史上最大IPO之一。Anthropic加入了一个拥挤的赛道。拥有xAI的SpaceX已提交上市文件，预计在六月完成。OpenAI据传也即将宣布自己的申请。在Anthropic提交文件后，我问OpenAI他们是否已提交或计划提交，对方回答："我们不予置评、确认、否认、猜测、高谈阔论或详细说明。"
Brian Barrett：哇。
Zoë Schiffer：这回应——
Leah Feiger：太绝了。
Zoë Schiffer：——非常特别。是的。
Brian Barrett：当他们搬出同义词词典时，你就知道事情严重了。
Zoë Schiffer：没错。我只回了句"哇！"，然后就已读不回了。所以，这确实是场全面竞赛，尽管OpenAI表示谁先上市无关紧要。但我认为，尤其在当前Anthropic势头正盛、公关表现极佳的节点，他们提交文件意义重大。但SpaceX、Anthropic、OpenAI要实现上市，还需完成大量准备工作。一切必须滴水不漏，所有环节需安排妥当。而这些公司仍是年轻甚至有时混乱的初创企业。
Brian Barrett：可怜的SpaceX，他们本不那么年轻、也没那么混乱，直到被迫接手xAI及其带来的麻烦。我确实为他们感到遗憾——他们本已达到所需标准，却在关键时刻陷入与其他AI公司相同的混乱。
Zoë Schiffer：确实如此。这个故事有许多有趣的支线，但有一件事会让你们觉得非常滑稽：旧金山已出现多套房产挂牌接受Anthropic或OpenAI股票作为购房款。这不是个例，而是多次出现。
Brian Barrett：难以置信。Zoë，你知道是否真有人用股票完成了购房交易吗？
Zoë Schiffer：不，疑问太多了。我最大的问题是：这些公司尚未上市，他们如何确定股票价值？另外，转让大量股票换房是否需要董事会批准？
Leah Feiger：老实说，感觉像是握手协议。不可能有多正式。
Zoë Schiffer：《连线》自由撰稿人Arielle Pardes就此写了篇文章。她问房产中介："客户预期这类交易如何操作？"所有人都含糊其辞："嗯，我们会想办法。"但一位中介表示已收到大量咨询，且声称主要来自Anthropic员工。有一套挂牌甚至明确只接受Anthropic股票而非OpenAI，因为房主更喜欢Anthropic的产品。
Brian Barrett：我认为有两方面：首先这是个不错的营销噱头——无论交易方式如何，我们都在讨论这些高价待售房产的地址；其次，这也提醒我们旧金山即将涌现更多亿万富翁——多到有人可能拿出价值200万美元的股票换房后仍剩余大量股票。这几乎像是虚拟货币了。
Zoë Schiffer：完全同意。很多人账面是百万甚至亿万富翁，但IPO后才会成为真正的富豪。有趣的是，在此期间他们已有机会将账面财富转化为实际房产价值。当然这是个大前提，因为我们尚未确认有这类交易实际发生。但就像Brian所说，最初看到几套挂牌时我觉得是噱头，但数量增多后我意识到事情不简单。当Ariel再次联系中介时，有些人已相当认真对待此事，不再是纯粹的营销手段。

Brian Barrett：你刚才提到，Zoë，Anthropic基本已表态：未经董事会批准出售股票的交易无效。这将会引发怎样的混乱，值得关注。
Zoë Schiffer：是的，我们一直看到员工在二级市场出售股票，但这通常不被建议。除非万不得已，否则不应这样做，因为价格会更低。有时人们确实需要变现，所以愿意这么做。但Anthropic暗示存在欺诈情况，并声称这些交易无效。
Leah Feiger：可怜的SpaceX和OpenAI还没享受这待遇。没房子出售，也没有遛狗换股票的交易。我等着看这一天。
Zoë Schiffer：OpenAI似乎也出现过几例，但SpaceX可能还没有。不过旧金山是个疯狂的地方，我不会惊讶。

Leah Feiger：我想聊聊AI行业另一件事。周二，特朗普总统签署行政令，要求科技公司在公开发布新AI模型前30天自愿向政府提供早期访问权限。数周来一直有行政令即将出台的报道，但特朗普在最后时刻否决原提案后，其最终落地曾存变数。以下是记者问及其理由时他的回应。
（特朗普档案音频）："我不喜欢其中某些方面，所以推迟了。我认为这妨碍——我们领先中国，领先所有国家，我不想做任何阻碍这种领先的事情。我们在AI领域有巨大优势，它带来巨大好处。我确实认为那可能成为障碍，我想确保它不会成为障碍。"
Leah Feiger：这是两周前他取消行政令时的表态。但这背后还有一点：政府内部有众多官员相互掣肘，将他拉向不同方向。我们不能忘记他的前AI沙皇David Sacks，正是他阻止了原行政令。最新签署的版本与原版无实质差异，主要区别在于：此前政府要求企业自愿提供90天提前访问权，而现在改为30天——显然各方都接受了这个期限。包括《连线》的Hugo Lowell和Maxwell Zeff在内的一些优秀报道指出，特朗普的高级助手说服他签署新行政令的理由是：鉴于AI技术已如此先进，他不能无限期推迟建立监管框架。
Zoë Schiffer：我喜欢这营销手法，抱歉。
Brian Barrett：他们不能像特朗普擅长的那样说"两周后再办"，最终确实需要行动。但Zoë，你为何喜欢这营销？
Zoë Schiffer：我只是觉得AI公司太擅长塑造形象了："我们如此强大、如此出色，以至于你们必须制定规则，因为我们即将爆发，局面会非常疯狂。"
Leah Feiger：百分之百，这就是事实。很多人问："AI公司愤怒吗？"不。
Zoë Schiffer：不。
Leah Feiger：他们私下正在与这些人会面。这是自愿的，完全没问题。
Zoë Schiffer：不过Hugo和Max文章中提到行政令的赢家与输家很有意思——苏茜·威尔斯似乎本轮获胜，而David Sacks未能如愿。他在行政令公布后在X上为其辩护，称自己支持该令，并将90天改为30天视为关键变量——这将产生巨大影响，因为AI公司现在能与政府同步协作而不会延迟产品发布周期。
Brian Barrett：即使是90天改30天，也是营销的一部分——"我们迭代速度太快，90天前的东西与30天前完全不同"，这或许有道理，但我们也看到这些公司多次迭代发布。实际上，这是在压缩审查时间——30天的窗口期短到无法进行任何有意义的审查。
Leah Feiger：再次强调，这完全是自愿的。称其为监管——
Zoë Schiffer：有些牵强。
Leah Feiger：对，我甚至不太愿意称之为监管。我们已经看到特朗普政府以多种方式利用这一点。Hugo和Max报道称，随着行政令生效，财政部长斯科特·贝森特正探索与中国的讨论，以建立共享高级AI系统信息的跨境框架。这很有趣。某种程度上这是个勾选框——我们得表明有所作为。这意味着什么？下一步如何走？政府已有一些AI审查流程，但审查、监管与"我只是想获取资源"是完全不同的概念，尽管听起来有些吹毛求疵。
Zoë Schiffer：不，这不是吹毛求疵，而是重要澄清。从行业角度看，与政府合作很重要。比如Anthropic与五角大楼的对抗——他们推出一个庞大到让情报机构必须深入了解的模型。据消息人士称，Anthropic与五角大楼的关系仍有些冷淡，但与情报机构的关系则更为融洽。我几乎将行政令解读为："让政府提前体验未来可能购买的产品。"就像"先试后买"，之后他们可能提供反馈，或认可其重要性并签订合同。
Brian Barrett：而且Anthropic的情况也说明，政府确实有手段迫使企业进入这个30天窗口。虽是自愿，但他们可以说："不提交模型审查的公司将无法获得合同。"那可是大笔资金和巨大权力。所以仍有操作空间。我同意这在很多方面是表演，但总有办法让其发挥作用。
Zoë Schiffer：是的，我认为大多数AI公司不会冒险不与政府合作，因为他们正努力成为政府承包商，同时也不希望未来面临更严格的监管。因此他们愿意在自愿层面合作，以免后续出现更严苛的规定。

Brian Barrett：说到网络安全（不过规模小得多），Meta近期遭遇了一系列令人紧张的黑客攻击。上周末，多名用户在社交媒体称Instagram账户被盗。其中一些账户知名度极高，包括奥巴马在白宫期间使用的认证账户。美国太空军总军士长（众所周知是约翰·本蒂维尼亚）的账户也遭到攻击。这些账户在被恢复前通常被灌入亲伊朗内容。这类黑客事件并不新鲜，但此次令人震惊之处在于黑客的作案手法：他们直接向Instagram的AI聊天机器人请求帮助获取账户访问权限。就是这么简单。X上甚至有视频演示了完整步骤。请勿在家尝试，千万别试。但基本上，黑客使用VPN伪装目标所在地，然后要求向该账户添加邮箱地址。AI聊天机器人回复"没问题"，向该邮箱发送确认邮件，然后账户就沦陷了。令人惊讶的是，居然没人进行故障排查，没预见到这种可能。
Leah Feiger：听起来这根本不该发生。太简单了，简直疯狂。
Zoë Schiffer：但这很有意思，它预示着随着大公司将原本由人工处理的客户服务工作转交给AI代理，可能会出现更多此类网络安全事件。当然，人类也并非无懈可击——我们见过无数SIM卡交换攻击，只因某人走进AT&T门店说"嘿，想搞点犯罪吗？"对方说"好啊，给钱就干"导致SIM卡被换。但当大量客服工作转交给AI代理时，系统性问题就会暴露——这些问题若有真人员工本可能更容易预防。
Brian Barrett：关键在于系统性——单个T-Mobile员工一次能造成巨大损害，但那只是一个人。而如果AI代理在这种条件下默认如此运作，你可以无限次重复操作，对任意账户持续攻击而毫无阻碍。尽管我们今天已讨论了许多关于Mythos模型、其网络安全影响以及发现漏洞的能力，但另一面是：AI代理可能相当愚蠢。
Zoë Schiffer：是的，不需要太复杂的操作。
Brian Barrett：你应该担忧AI有多聪明，也该担忧AI有多愚蠢——这就是问题的两面性。
Leah Feiger：往小处想吧，朋友们。这并非最严重的黑客事件。Meta表示已修复问题，但我想到所有其他网站和AI系统面临的数据安全风险——越来越多功能强大的黑客攻击正瞄准这些系统。从转录服务到各类网站，一切都在转向AI代理化。我很关注这个问题将如何解决。其他公司是否因Meta的重大失误而陷入恐慌？我不知道。
Brian Barrett：我认为不足以让他们放弃原有计划。这将是典型的"亡羊补牢"——在有人被撞后才设置路障。这种情况会反复出现，因为无论内部测试和红队演练（即模拟黑客攻击测试）多么完善，总会有人想到新方法。Meta只是一家公司，面对的是全世界的变态、怪胎和破坏分子——人数众多。
Zoë Schiffer：变态和怪胎——我们的播客新名字。说到红队演练，这通常是信任与安全团队的工作。而这些团队——
Leah Feiger：现在都没了。
Zoë Schiffer：——规模已大不如前。所以看看事态如何发展会很有趣。Meta内部，我们本周与一些人聊过，他们对这消息只能叹气。公司刚裁掉大量员工，我们报道过此事。我问他们："现在情况如何？"这次黑客事件成了与他们交流的借口。他们回答："如你所料，我们现在一个人干两个人的活。"你能想象后果如何。
Brian Barrett：我们刚才讨论AI监管时，重点都在国家安全等宏观层面，但消费者端产品受到的关注不足。如果存在某个监管消费者金融保护机构，此刻就会很有帮助。我们曾有过这样的机构，理论上现在也有，但实际并非如此。这种全面放松监管的现状，恰逢原有工具失效、新工具又极其不可靠。我们将看到更多此类事件。

Leah Feiger：我能换个与AI无关的话题吗？
Brian Barrett：请便。
Zoë Schiffer：哇，我以为不存在这种话题呢。说吧，女王。
Brian Barrett：而且我们大概也能设法扯回AI。
Zoë Schiffer：能，我们能。
Leah Feiger：不，绝对扯不上。好吧，这个故事我们长期关注、报道和审视，关于一名DOGE举报人刚对埃隆·马斯克提起诉讼。一切始于去年。2025年4月14日，国家劳资关系委员会IT职员丹·贝鲁利斯提交举报，声称DOGE已危及该机构数据安全并疑似外泄。
（档案音频）"一名举报人声称，DOGE不仅访问了其所在机构的数据，还带走大量敏感信息。根据提交国会的披露，约10GB数据——相当于印刷成纸质文档后一整套百科全书的体量。"
Leah Feiger：这一指控非同小可，尤其当时DOGE团队正解雇联邦雇员、全美范围内访问敏感数据。去年四月正是高峰期。贝鲁利斯通过NPR文章公开身份，声称有人将恐吓信贴在他门上，他本就因公开言论而恐惧。时间快进，贝鲁利斯现已在华盛顿特区法院起诉马斯克诽谤。他声称马斯克谎称他的举报是假的，从而使他成为暴力目标。这一指控极其激烈，背后的背景是：马斯克去年转发了一名右翼网红的X帖子，声称DOGE已洗清嫌疑，该举报人的证词纯属捏造。此后——
Brian Barrett：AI部分要来了。
Leah Feiger：不，绝对没有。此后，贝鲁利斯称他驾车时刹车被剪断，他认为与此相关。这就是诽谤诉讼的核心，相当重大。举报人称马斯克在X上的"垃圾帖"可能致其受伤。
Brian Barrett：这听起来很极端，我本能地对这类指控持怀疑态度，因为它像电影情节。但诉讼文件中包含确凿证据：修车技师证明不仅刹车被剪断，安全气囊也被破坏，汽车电路被改造导致气囊传感器失效。事发前不久有人将恐吓信贴在他家门上——证明有人知道他的住址并特意威胁他。这些都是可验证的惊人事实。显然马斯克拥有巨大平台，当他发表煽动性言论时，人们会当真。
Zoë Schiffer：是的。读这篇文章时，我立刻想到两起历史事件。一是马斯克转发暗示Twitter前信任与安全主管约尔·罗斯是恋童癖的帖子，导致罗斯不得不与丈夫离开住所、躲藏起来，最终因死亡威胁出售房屋。约尔未起诉马斯克，但多年前另一起事件中有人起诉——不知道你们是否记得"恋童癖潜水员"事件？
Leah Feiger：天哪，记得。
Brian Barrett：那个洞穴潜水员。
Zoë Schiffer：第三个例子。正是如此：马斯克暗示参与救援泰国少年足球队的洞穴潜水员是恋童癖。那名男子在英国提起诉讼，结果马斯克胜诉。
Leah Feiger：我好奇这件事是否对马斯克有所触动。他可能觉得："我可以随时发垃圾帖。"但确实有迹可循。我永远记得去年DOGE事件期间，他疯狂发帖、发表吓人的煽动性言论。《连线》记者David Gilbert不得不撰写辟谣文章，澄清没有百岁老人领取社保福利——因为马斯克分享的虚假声明引发民众致电社保办公室质问："怎么回事？我的税金都花哪去了？"他的行为有直接后果，我好奇他是否真的需要吸取教训。
Brian Barrett：我怀疑他不会。就连美国国际开发署的案例——他散布大量阴谋论，然后裁撤该机构，导致全球性损害。
Leah Feiger：至今仍在持续。
Zoë Schiffer：唯一让他尝到苦头的是SEC那次——他在毫无证据且明显虚假的情况下声称"已获得资金"将特斯拉私有化。
Brian Barrett：以每股420美元？
Leah Feiger：哦，没错。
Zoë Schiffer：对，我知道有个大麻梗。他最后好像交了2000万美元罚款？
Brian Barrett：对他来说是零花钱，无论数额多少都相当少。
Leah Feiger：这位下周可能成为万亿富翁的男人。
Zoë Schiffer：没错。他谈及此事时确实流泪了，显然对此很不满。但这事发生在我们讨论的这一切之前，所以感觉他并没有吸取多大教训。
Leah Feiger：我很好奇此案走向。历史上美国举报人保护制度相当强大，但在特朗普2.0时期情况有所不同。这案子相当离奇。请前往WIRED.com阅读我们的记者Vittoria Elliott的报道，并告诉我们你的看法。

Zoë Schiffer：今天的节目到此结束。我们将在节目说明中附上所有提及文章的链接。《Uncanny Valley》由Kaleidoscope Content制作。Adriana Tapia负责本期制作，Amar Lal在Macro Sound完成混音，Daniel Roman担任事实核查，Pran Bandi是纽约录音室工程师，Marc Leyda是旧金山录音室工程师，Kimberly Chua是高级数字制作经理，Kate Osborn是执行制作人，Katie Drummond是《连线》全球编辑总监。

英文来源：

This week on Uncanny Valley, we discuss why the race among the top AI companies to go public has just gotten started but is already creating bizarre repercussions—including numerous San Francisco real estate listings preferring Anthropic stock instead of cash. They also get into why Donald Trump’s new executive order on AI safety is underwhelming and how hackers were able to use Instagram’s AI chatbot to access high-profile accounts like President Barack Obama’s. Plus, we report on how a DOGE whistleblower is suing Elon Musk for defamation after the billionaire publicly called him a liar—and the next thing he knew, his life was in danger.
Articles mentioned in this episode:

• What’s Worth More Than Cash in San Francisco Real Estate? Anthropic Stock
• This Is How Trump Finally Signed the AI Executive Order
• He Blew the Whistle on DOGE. Then His Brakes Were Cut
  You can follow Brian Barrett on Bluesky at @brbarrett, Zoë Schiffer on Bluesky at @zoeschiffer, and Leah Feiger on Bluesky at @leahfeiger. Write to us at [email protected].
  How to Listen
  You can always listen to this week's podcast through the audio player on this page, but if you want to subscribe for free to get every episode, here's how:
  If you're on an iPhone or iPad, open the app called Podcasts, or just tap this link. You can also download an app like Overcast or Pocket Casts and search for “uncanny valley.” We’re on Spotify too.
  Transcript
  Note: This is an automated transcript, which may contain errors.
  Brian Barrett: Hey, this is Brian. Before we start, two quick things. If you've been enjoying listening to the show, we would appreciate it if you took a second to rate it in your podcast app of choice. It really helps us reach more people. And second, if you have any questions related to tech, privacy, or politics that you would like me, Zoë, and Leah to take on, now is the time to submit them to [email protected]. It doesn't matter how big or how small, we want to hear from you and get you answers. OK, on to the show.
  Zoë Schiffer: Hi guys.
  Brian Barrett: Hi everybody. We are far-flung today.
  Zoë Schiffer: I know. You're not even supposed to be here, but you're here.
  Brian Barrett: Couldn't miss it. I am out this week, technically. I'm at the beach with my parents and my family, and we're playing a lot of mini golf. That's basically it.
  Leah Feiger: That's pretty good stuff though. Who's winning?
  Brian Barrett: Grandma. Grandma crushing it.
  Zoë Schiffer: Wow.
  Brian Barrett: A terror on the links.
  Zoë Schiffer: Welcome to WIRED's Uncanny Valley. I'm Zoë Schiffer, director of business and industry.
  Brian Barrett: I'm Brian Barrett, executive editor.
  Leah Feiger: And I'm Leah Feiger, director of politics and science.
  Zoë Schiffer: Today on the show, we're diving into the IPO bonanza that the top AI companies are embarking on to the point where some real estate listings in San Francisco are looking for not just regular old cash but something that is considered even more valuable right now: stock in Anthropic and OpenAI. The madness is honestly just beginning.
  Leah Feiger: And this week the Trump administration announced an executive order asking AI companies to give the federal government access to their most advanced artificial intelligence models 30 days before their public release. This decision did not come that smoothly. We're going to discuss all the inside fighting that has been going on inside the administration when it comes to AI regulation and whether this executive order actually makes any difference.
  Brian Barrett: We'll also get into how hackers tricked Instagram's AI chatbot to access some high-profile accounts, including President Obama's former White House account. And we'll talk about a DOGE whistleblower who says someone cut the brakes of his car a few days after Elon Musk boosted a claim that called him a liar.
  Zoë Schiffer: So you've probably already heard this week that Anthropic officially submitted its paperwork to go public. It did this confidentially, and then it shouted it from the rooftops.
  Brian Barrett: And they beat OpenAI.
  Zoë Schiffer: And they beat OpenAI, which we will get into in just a moment. The company has not set the exact date or the specific amount of money it's seeking to raise, but with its current valuation, which is a mind-boggling $965 billion, that's billion with a B. It is well on its way to becoming one of the largest IPOs in history. Anthropic is joining a pretty crowded field. SpaceX, which owns xAI, already filed its paperwork to go public. That's probably happening in June. And OpenAI is also rumored to be announcing their own filing very, very soon. I asked OpenAI to weigh in after Anthropic filed and kind of whether they had already filed, were planning to, and they said, "We are not going to comment, confirm, deny, speculate, pontificate, or elaborate."
  Brian Barrett: Wow.
  Zoë Schiffer: Which was—
  Leah Feiger: That's a great response.
  Zoë Schiffer: —so special. Yeah.
  Brian Barrett: When they bust out the thesaurus, you know it's serious.
  Zoë Schiffer: Yeah. I just responded, "Wow!" and that was left on read. So yeah, I mean, this is a full-on race, though OpenAI has said it doesn't matter who goes public first. I think particularly in this moment, when it feels like Anthropic is really ascendant and at least from a PR perspective, they're doing incredibly well. I think it matters that they have filed, that they're well on their way. But there's also a lot of stuff that needs to happen for SpaceX, for Anthropic, for OpenAI, for them to be able to go public. You need to be super buttoned up. You need to have all of your ducks in a row. And these are young and at times messy startups.
  Brian Barrett: Well, and poor SpaceX, which is not as young and was not as messy until they sort of had a forced adoption of xAI and all that goes with that. So I do feel bad for them that they were all of the things that you need to be until right up to the moment that they needed to be them, and now they're just in the same AI mess as everybody else.
  Zoë Schiffer: Yeah, they really, really are. So there've been a lot of interesting sub-threads to this story, as you might imagine. But one thing that I think you will both find very, very funny is that we have been seeing real estate listings in San Francisco say that they will accept stock in Anthropic or OpenAI in exchange for a house. And this has happened not just one time—multiple times.
  Brian Barrett: Incredible. Zoë, do we know if anyone has actually done a purchase like this yet or is that still—
  Zoë Schiffer: No, there are so many questions. My major one was how are they deciding the value of the—these aren't public companies. Also, do you need board approval to give away a bunch of your stock in exchange for a house?
  Leah Feiger: It feels like it's a handshake agreement to be honest. There can't be anything that official here.
  Zoë Schiffer: WIRED freelancer Arielle Pardes wrote a story for us about this, and she asked the real estate agents, “How are your clients expecting these deals to go?” And everyone was kind of like, “No, we'll figure it out.” But one of the agents did say that they've received a lot of interest, and she claims specifically from Anthropic employees. There was one listing in particular that said they were only accepting Anthropic stock, not OpenAI, because the guy who was selling the house likes Anthropic’s products better.
  Brian Barrett: I think a couple things. One, it's also just a good marketing stunt—we're talking about and listing the addresses of these very expensive houses for sale, regardless of how they buy them. It's just good on these realtors for getting the word out there. But also it is a reminder of how many billionaires are about to hit the streets in San Francisco, even more than there already are to the point where, I mean, people can probably, I assume a lot of people at Anthropic or OpenAI who can give away $2 million worth of stock and still have a whole lot of stock left over. It's sort of like monopoly money at this point.
  Zoë Schiffer: Totally. I mean, a lot of these people like you insinuated are on paper millionaires or even billionaires, but once the IPOs happen, there'll be actual millionaires, if not billionaires. It's interesting that even in the meantime, there's potential for them to convert their paper wealth into actual value in the form of a house. That's like a big if, because again, we haven't seen one of these transactions play out that we actually know of. But like you said, Brian, when we first saw a couple listings, I was like nah, this feels like a gimmick. But then enough of them popped up that I was like, OK, there's something here. And when Ariel reached out to speak to the agents again, some of them were taking it pretty seriously. It didn't feel like it was a pure marketing play, even if it does have that effect.
  Brian Barrett: You hinted at this earlier, Zoë, but Anthropic has basically said that if you try to sell your Anthropic shares without board approval, then that transaction is not valid, right? It is going to be interesting to see what kind of messes—
  Zoë Schiffer: Yeah. I mean, we have always seen, I think, employees selling stock on secondary markets. It's usually not advised. You don't want to do that, because you're going to get a worse price unless you really, really have to. And sometimes you do have to. Sometimes people need liquidity, and so they're willing to do it. But Anthropic I think has insinuated that there's some scammy situations going down and that those transactions are not valid.
  Leah Feiger: Poor SpaceX and OpenAI, that's not happening yet. No houses for sale, no exchange of dog walking for stock. I'm waiting. I'm waiting for the moment.
  Zoë Schiffer: I think OpenAI has seen a few, but yeah, maybe no SpaceX quite yet. Although San Francisco’s a crazy place, so I wouldn't be surprised.
  Leah Feiger: I want to talk about something in the AI industry that is not having to do with IPOs. On Tuesday, President Donald Trump signed an executive order that asked tech companies to voluntarily give the government early access to new AI models before releasing them to the public, 30 days before they're set to be released to be exact. There had been reports of an executive order coming down the pipeline for weeks now, but after President Trump nixed the original proposal at the very last minute a few weeks ago, it wasn't clear if it was actually going to happen. Here's his response when he was asked about his reasoning.
  Donald Trump, archival audio: I didn't like certain aspects of it. I postponed it. I think it gets in the way of—we're leading China, we're leading everybody, and I don't want to do anything that's going to get in the way of that lead. We have a very substantial on AI. It's causing tremendous good. I really thought that could have been a blocker, and I want to make sure that it's not.
  Leah Feiger: He said all of those things when he canceled the executive order a couple of weeks ago now. But the other part of this is there were so many competing officials in the administration that were pulling him in different directions. We can't forget David Sacks, his former AI czar in all of this, who very much stopped the original executive order from happening. While the most recent signed order doesn't have hugely substantial differences, the big difference that we can point to is, beforehand the government was asking for 90 days for companies to voluntarily give, as opposed to the 30 days which everyone apparently is now comfortable with. There's been some really good reporting out there, including from WIRED's Hugo Lowell and Maxwell Zeff, who reported just on Wednesday that Trump's senior aides actually persuaded him to sign this new executive order by saying that he could not infinitely delay establishing a regulatory framework for the tech, considering that it was just so advanced at this point.
  Zoë Schiffer: I love all the marketing, sorry.
  Brian Barrett: They can't do that sort of “we'll do it in two weeks” that I think Trump is so good at. You really do eventually need to get it. But Zoë, what do you love about the marketing?
  Zoë Schiffer: I just feel like the AI companies do such a good job of being like, "We're actually so good at what we do and so powerful that you're like, you need to set some ground rules because we're about to explode. It's going to be so crazy."
  Leah Feiger: It's 100 percent. That's 100 percent what happened. I mean, so many people are like, "Are they furious? Are the AI companies?" No.
  Zoë Schiffer: No.
  Leah Feiger: They're behind the scenes. They're having meetings with all of these folks. This is voluntary. This is 100 percent fine.
  Zoë Schiffer: I did think it was really interesting in Hugo and Max’s article where they talked about the winners and the losers with the executive order, because it does seem like Susie Wiles has kind of won this round, whereas David Sacks, the former AI and crypto czar, as you mentioned, did not get his full wish. He went on X shortly after this announcement dropped and kind of defended the order, tried to say he was behind it and really framed it as like the 90- to 30-day switch that we saw in the planned order and what actually was signed and came out was kind of the X factor for him, and it was going to make a huge difference because now AI companies could work in lockstep with the government and not delay their product launch cycle.
  Brian Barrett: Even the 90- to 30-day thing is part of that market—it is of a piece of the marketing thing, right? It's like we are iterating so quickly, and we are so at the cutting edge. Ninety days is insane. Whatever you're looking at 90 days in advance won't be anything like what the 30 days in advance looks like, which there's probably an element of truth to that, but also we've seen a lot of iterative releases from these companies. I think really what it is is squeezing the amount of review time—to the extent that there is any actual review in this 30-day window—to be too short to really be meaningful in any way.
  Leah Feiger: And again, to be clear, this is all voluntary. There's calling this regulation is—
  Zoë Schiffer: A stretch.
  Leah Feiger: Yeah, I don't think I feel comfortable even really calling this a regulation. Look, already we're starting to see in so many ways that the Trump administration is already utilizing this to their benefit. Hugo and Max reported that now, with the order in place, Secretary of the Treasury Scott Bessent is now exploring discussions with China to figure out about creating a similar cross-border framework for sharing advanced AI system information. And that's really interesting to me. So in some ways this is sort of a check mark. We had to say that we're doing something on this. What does that mean, and how do we really go forward from there? And there are already some review processes for AI in the government. But again, review versus regulation versus, like, “I just want access to the goods” is all very different things, as nitpicky as that sounds.
  Zoë Schiffer: No, no, I think that's not a nitpick. It's important clarification. I mean, I think that from the industry's perspective, this type of working with the government is important. I'm thinking about the Mythos thing where Anthropic was in this kind of head-to-head fight with the Pentagon, then it comes out with a model that's so big and scary that intelligence agencies need to be really read in. And while Anthropic and the Pentagon relations I've heard from sources are still a little bit icy, Anthropic and intelligence agencies: a little warmer, a little warmer. So I think I was almost reading the executive order being like, “Oh, they're giving the government an early taste of what they could later buy.” They're like, “Check it out, try it before you buy it.” And then maybe they'll weigh in, maybe they'll have feedback, maybe they'll just be like, “This is really important, and we need to actually use it and we see a contract later.”
  Brian Barrett: And I'd say too, I think the Anthropic situation also speaks to, I mean, the government does have some levers to force people to get into this 30-day window. It's voluntary, but they could say, "We're not going to get contracts to anybody who doesn't actually submit their thing for review, submit their models for review." And that's a lot of money. It's a lot of power. So there are things they can do. I agree that it's all for show on both sides in a lot of ways, but there are ways to make it work.
  Zoë Schiffer: Yeah, no, I think most AI companies would be fearful to not play ball here, because they are all trying to quite literally work with the government, become government contractors, and they also just—they don't want harder regulation. So they want to play ball on this voluntary level so that more stringent regulation doesn't come down later.
  Brian Barrett: Speaking of cybersecurity, but on a much smaller scale, Meta dealt with a pretty stressful series of hacks recently. This past weekend, several users on social media claimed that their Instagram accounts had been compromised. Some of these accounts were pretty high-profile, including a verified account used by President Obama when he was in the White House. The account of the US Space Force's chief master sergeant, who we all know is John Bentivegna, also targeted in the hacking. These accounts were typically flooded with pro-Iran content before they were recovered. Now hacking incidents like this are not new. They happen all the time, but what was interesting and especially alarming in this case is how the hackers had done it, which is basically they asked Instagram's AI chatbot to help them get access. It's really basically that simple. There's a video on X that even shows the step-by-step process. Please don't try it at home, please don't do it. But basically the hacker used allegedly a VPN to spoof the presumed location of their target and then went ahead and just asked to add an email address to that account. The AI chatbot would say “sure,” send a confirmation email to that email address, and all of a sudden you've got access. It's pretty remarkable that no one did some troubleshooting here, maybe thought that that was possible.
  Leah Feiger: It sounds like something that should have not happened. It's so simple you guys. This is crazy.
  Zoë Schiffer: Yeah, but I mean, this is interesting because it says something about the kinds of cybersecurity incidents we might see more of as more big companies offload the work that was previously done—the customer experience work by humans to AI agents. And I want to say that humans aren't infallible. We've seen many a SIM swap take place because someone walked into an AT&T store and said, "Hey, do you want to get on on crimes?" And the person said, "Sure, we'll do that for money." And thus a SIM swap happened. But of course, when you're transferring a lot of customer experience work to agents, you can see systematic problems play out that might have been slightly more preventable with a human staff.
  Brian Barrett: It's the systematic part, because one T-Mobile employee one time can do a lot of damage, but it's one person. Whereas if this is just how the agent works under these conditions, you can do it again and again and again as many times as you want across as many accounts as you want with no stop. I do think for all of that we've already talked about today and all the concern and gnashing of teeth over Mythos and the cybersecurity implications of these models and how they can do these very sophisticated findings of bugs, there's the other end of it too, which is that AI agents can be pretty dumb.
  Zoë Schiffer: Yeah, you don't need to be that sophisticated for these.
  Brian Barrett: You should be worried about how smart AI is and how dumb AI is, I guess is the point. It's on both ends of the spectrum.
  Leah Feiger: Dream small, folks. It's not actually the most intense hacking moment. And look, I mean, Meta has said that they have since resolved the problem, but obviously I'm thinking about all of these other websites out there and all of these other concerns about the impact of increasingly capable and common hacking attempts on AI systems that have to do with people's data and security. Everything is switching over. I'm thinking of transcription services, websites. It's just so much. It's everything now and everything is AI-agent-oriented. I'm really interested to see how this gets addressed. Are other companies freaking out right now following Meta's big gaffe? I don't know.
  Brian Barrett: I think not enough to go back on any plans. I think it is going to be a classic scenario of you put up a stop sign after someone gets run over. I think that's going to be what we're going to see over and over again, because as much as you test internally, as much as you red-team things— red team is the term for when you sort of put on your hacker hat and say, "If I were a bad guy, what would I do with this?" As much as you do that, there are going to be things that people think about. At Meta, you are one company, it's you versus the entire world of deviants and freaks and people who just want to break things, which is a lot of people.
  Zoë Schiffer: Deviants and freaks, the new name of our podcast. And this is, I mean, just going back to red teaming, that's work that a trust and safety team typically does. And those teams—
  Leah Feiger: We don't have those anymore.
  Zoë Schiffer: They're not as big as they used to be. There's just not as much work. So yeah, I mean, it will be interesting to see how this plays out. Obviously within Meta, we've been talking to folks this week who kind of met the news with a sigh. The company has just laid off a large portion of the workforce. We've written about that. We've talked about that. And I checked in with people being like, "Well, how's it going now?" The hack was kind of an excuse to talk to people, see how they're doing. And they're like, "I mean, as you'd expect, we're asked to do two jobs now instead of one." So you can imagine how that's playing out.
  Brian Barrett: I also, we were talking about AI regulation earlier and all this emphasis on national security and these high-level things, but again, not as much on consumer-facing products, which would be if you had say some sort of bureau that looked after consumer finances and protecting that, that would be helpful to have in this moment as well. We used to have one of those. Technically, I guess we still do. Not really. So all of this broader deregulation is coming at this moment when the tools that were once available are not. These new tools are very fallible. We're going to see a lot more of this.
  Leah Feiger: Can I bring us to a topic that has nothing to do with AI, guys?
  Brian Barrett: Please.
  Zoë Schiffer: Wow. I didn't know one existed, but yes, go off, queen.
  Brian Barrett: Also, I think we can probably try to find a way to tie it back in.
  Zoë Schiffer: We can. We can.
  Leah Feiger: No, absolutely not. Well, OK. This story is something that we have been thinking about, covering, looking at for a long time, but it is all about a DOGE whistleblower who just filed a lawsuit against Elon Musk. This all really started last year. On April 14, 2025, Dan Berulis, an IT staffer at the National Labor Relations Board, the NLRB, filed a whistleblower complaint with a massive claim. He said that DOGE had compromised the agency's data and appeared to be exfiltrating it out of the NLRB.
  Archival audio: A whistleblower is coming forward with claims that DOGE not only accessed data from his agency but also took a substantial amount of sensitive data with them. According to a disclosure shared with Congress, “Around 10 gigabytes of data, the equivalent of a full stack of encyclopedia is worth if someone printed these files as hard copy documents.”
  Leah Feiger: This was a massive claim, especially at the same time as you guys very much remember, DOGE teams were firing federal workers and accessing sensitive data across the country. We were in the height of this last year in April. Berulis went public in an NPR article. His name was attached to it, and he claimed a threatening note had been taped to his door, and he was already scared about speaking out. Fast-forward a little bit, Berulis has now filed a defamation lawsuit in a DC court against Elon Musk. He said that Musk made him a target of further violence by falsely stating that Berulis' whistleblower claim against DOGE was fake. This is a really intense claim for a variety of reasons, and what this all really harkens back to is Musk last year resharing an X post from a right-wing influencer claiming that DOGE had been cleared and that this whistleblower's testimony was fake basically. After that happened—
  Brian Barrett: Here comes the AI part.
  Leah Feiger: No, absolutely not. After that happened, Berulis says that he was driving his car and his brakes were cut, and he believes that it is related. So that is the crux of this defamation lawsuit against Elon Musk, and it's kind of a big one. This is a whistleblower who says that Elon Musk's shitposting on X could have caused him injury.
  Brian Barrett: And it sounds so extreme, and I'm always a little bit inherently skeptical of claims like this, because it sounds like it's out of a movie, but there are just some facts that are in part of this suit that he filed. He has the actual documentation from the person who worked on his car who said that not only were the brakes cut, but the airbags were basically made not to work and that the car had been rewired so that the sensors wouldn't go off to say that the airbags had been disabled. He had received a threatening note on his door not long before this happened, so someone knew where he lived and had gone out of their way to threaten him. So there's a lot of stuff here that is verifiable that's really alarming. And I think obviously Elon Musk has a huge platform, and when he says inflammatory things, people listen.
  Zoë Schiffer: Yeah. I mean, I had two kind of historical incidents instantly come to mind when I was reading this piece. One was when Elon Musk reposted something on X insinuating that Yoel Roth, the former head of trust and safety at Twitter, was essentially a pedophile, and Yoel had to leave his house where he lived with his husband, basically go into hiding. And they eventually sold that house because of the death threats that they received. Yoel never filed a lawsuit against Elon Musk due to this incident, but someone else did years prior. I don't know if you guys recall the pedo-guy incident. This was—
  Leah Feiger: Oh my god. Yeah.
  Brian Barrett: The cave diver.
  Zoë Schiffer: —the third example. Exactly. Of Elon Musk insinuating that one of the cave divers that helped save the team of Thai soccer players, the young young boys, was a pedophile. That man sued, I believe this was in the UK, and Elon Musk was actually victorious in that suit. The guy did not win the defamation case against him.
  Leah Feiger: I mean, I'm interested if that stayed with him at all. He was like, “I can actually just shitpost anytime I want.” There is something there, though. I mean, I'll never forget last year during DOGE, everything, he was posting up a storm. He was making inflammatory comments all the time that really freaked people out. David Gilbert, one of WIRED's reporters, had to do an entire debunker piece about know how no, 100-plus-year-olds were not filing for Social Security benefits, because Elon Musk was sharing claims to the contrary. And as a result, everyone started calling Social Security offices being like, "What the hell? This is where my tax dollars are going?” His actions have direct consequences, and I'm curious if he's actually going to have to learn a lesson from this.
  Brian Barrett: I doubt it. I mean, even USAID, right? He texted so many conspiracies theories around USAID and then gutted the agency, which led to huge harms all over the world.
  Leah Feiger: And is still.
  Zoë Schiffer: But the one time that this has come back to bite him a little bit was with the SEC when he claimed without evidence and apparently falsely that he had funding secured to take Tesla private.
  Brian Barrett: At $420 a share, Zoë?
  Leah Feiger: Oh, yeah.
  Zoë Schiffer: Yeah. I was like, I know that there was a weed joke. And I think he had to pay, what was it, like a $20 million fine or something?
  Brian Barrett: It was pocket change for him, whatever it was. It was pretty small. Yeah.
  Leah Feiger: The man who's set to become a trillionaire next week, possibly.
  Zoë Schiffer: Yeah, yeah, yeah. He did tear up talking about it. I think he really did not appreciate this incident, but that happened before everything else we're talking about. So it doesn't feel like he's learned an enormous lesson.
  Leah Feiger: I'm really curious to see where this one goes. Obviously, whistleblower protections are historically strong in this country. Things are a little bit different in Trump 2.0, but it's a pretty wild one. Again, from our reporter, Vittoria Elliott, go give it a read on WIRED.com. Let us know what you think.
  Zoë Schiffer: That's our show for today. We'll link to all the stories we spoke about in the show notes. Uncanny Valley is produced by Kaleidoscope Content. Adriana Tapia produced this episode. It was mixed by Amar Lal at Macro Sound. It was fact-checked by Daniel Roman. Pran Bandi is our New York studio engineer. Marc Leyda is our San Francisco studio engineer. Kimberly Chua is our senior digital production manager. Kate Osborn is our executive producer, and Katie Drummond is WIRED’s global editorial director.