I don't think Sam is the guy who should have his finger on the button.

我不认为山姆是那个应该把手指放在按钮上的人。

— Ilya Sutskever, OpenAI Chief Scientist, to a fellow board member, 2023

— 伊利亚·苏茨克维，OpenAI首席科学家，对一位董事会成员说，2023年

Sam Altman surrounded by masks of his own face — visual by David Szauder, generated using AI

Visual by David Szauder; Generated using A.I. — The New Yorker

视觉设计：David Szauder；AI生成 — 《纽约客》

⚡ THE NEW YORKER INVESTIGATION

⚡ 《纽约客》深度调查

Sam Altman May Control Our Future — Can He Be Trusted?

山姆·奥特曼可能掌控我们的未来——他值得信任吗？

New interviews and closely guarded documents shed light on the persistent doubts about the head of OpenAI. Based on the April 2026 investigation by Ronan Farrow & Andrew Marantz.

全新采访与严密保管的文件，揭开了围绕OpenAI掌门人挥之不去的信任疑云。基于罗南·法罗与安德鲁·马兰茨2026年4月的深度调查报道。

Read the full article on The New Yorker → 阅读《纽约客》原文 →

Key Players

关键人物

Sam Altman

山姆·奥特曼

CEO, OpenAI

OpenAI首席执行官

Former Y Combinator president. Co-founded OpenAI with Musk in 2015. Fired by the board in Nov 2023, reinstated five days later. Driving a potential $1 trillion IPO.

前Y Combinator总裁。2015年与马斯克共同创立OpenAI。2023年11月被董事会解雇，五天后复职。正推动可能达万亿美元估值的IPO。

Ilya Sutskever

伊利亚·苏茨克维

Former Chief Scientist, OpenAI

OpenAI前首席科学家

Born in the Soviet Union. Called the most gifted AI scientist of his generation. Compiled the legendary ~70-page "Ilya Memos" documenting Altman's alleged deceptions. Resigned after the Blip; co-founded Safe Superintelligence.

生于苏联。被誉为同辈中最有天赋的AI科学家。编纂了传奇的约70页"伊利亚备忘录"，记录奥特曼涉嫌欺骗的行为。"闪电事件"后辞职，联合创立了Safe Superintelligence。

Dario Amodei

达里奥·阿莫代

Co-founder & CEO, Anthropic

Anthropic联合创始人兼CEO

Biophysicist. Led OpenAI's safety team. Kept 200+ pages of private notes titled "My Experience with OpenAI." Left in 2020 to co-found Anthropic with his sister Daniela. Wrote: "The problem with OpenAI is Sam himself."

生物物理学家。曾领导OpenAI安全团队。保存了200多页名为"我在OpenAI的经历"的私人笔记。2020年与妹妹达妮埃拉离开，联合创办Anthropic。曾写道："OpenAI的问题就是山姆本人。"

Greg Brockman

格雷格·布罗克曼

Co-founder & President, OpenAI

OpenAI联合创始人兼总裁

Altman's second-in-command. His diary entries reveal conflicted instincts: "Happy to not become rich on this" alongside "What will take me to $1B." Committed $50M to a pro-AI super PAC. His stake reportedly worth ~$20 billion.

奥特曼的副手。日记暴露了矛盾心理："乐意不从中致富"与"什么能让我达到10亿美元"并存。向亲AI超级政治行动委员会投入5000万美元。据报道持股价值约200亿美元。

Elon Musk

埃隆·马斯克

Co-founder, OpenAI (departed 2018)

OpenAI联合创始人（2018年离开）

Provided early funding and office space. Quit acrimoniously after demanding majority control. Founded rival xAI in 2023. Sued Altman for fraud the following year. Calls him "Scam Altman" and "Swindly Sam" on X. Intermediaries connected to Musk have circulated extensive opposition research.

提供了早期资金和办公场所。在要求获得多数控制权后愤然离开。2023年创立竞争对手xAI。次年以欺诈罪起诉奥特曼。在X上称其为"骗子奥特曼"和"花言巧语山姆"。与马斯克有关的中间人散布了大量反对派调查材料。

Mira Murati

米拉·穆拉蒂

Former CTO, OpenAI

OpenAI前首席技术官

Served as interim CEO during the five-day firing. Gave Sutskever material for his memos. Eventually signed the letter demanding Altman's return. Left in 2024 to build her own AI startup. "We need institutions worthy of the power they wield."

在五天解雇期间担任临时CEO。向苏茨克维提供了备忘录材料。最终签署了要求奥特曼回归的信。2024年离职创办自己的AI公司。"我们需要配得上其权力的机构。"

Helen Toner

海伦·托纳

Former Board Member, OpenAI

OpenAI前董事会成员

AI policy expert. Received Sutskever's memos as confirmation of what she already suspected. Lost her board seat after Altman's reinstatement. "Control Z — or the company falls apart."

AI政策专家。收到苏茨克维的备忘录，印证了她早已有的怀疑。奥特曼复职后失去董事席位。"撤销——或者公司崩溃。"

Jan Leike

扬·莱克

Former Superalignment Lead, OpenAI

OpenAI前超级对齐团队负责人

Co-led the superalignment team with Sutskever. Emailed the board warning OpenAI was "going off the rails on its mission." Resigned when safety culture eroded. Wrote on X: "Safety culture and processes have taken a backseat to shiny products."

与苏茨克维共同领导超级对齐团队。给董事会发邮件警告OpenAI正在"偏离使命轨道"。安全文化遭侵蚀后辞职。在X上写道："安全文化和流程已让位于光鲜的产品。"

Paul Graham

保罗·格雷厄姆

Co-founder, Y Combinator

Y Combinator联合创始人

Altman's early mentor. Recruited him to run Y.C. at twenty-eight. Publicly maintained Altman left voluntarily. Privately told colleagues: "Sam had been lying to us all the time."

奥特曼的早期导师。在他28岁时招募他运营Y.C.。公开表示奥特曼是自愿离开。私下对同事说："山姆一直在对我们撒谎。"

Satya Nadella

萨蒂亚·纳德拉

CEO, Microsoft

微软CEO

Invested ~$13B in OpenAI. Blindsided by the firing. Coordinated closely with Altman during the Blip to engineer his return. Senior Microsoft executives now say: "He has misrepresented, distorted, renegotiated, reneged on agreements."

向OpenAI投资约130亿美元。对解雇事件毫不知情。在"闪电事件"期间与奥特曼密切协调以策划其回归。微软高管现在说："他歪曲事实、扭曲协议、反复无常。"

Brian Chesky

布莱恩·切斯基

Co-founder, Airbnb

Airbnb联合创始人

Altman's fiercest loyalist. Joined the "war room" at Altman's mansion during the Blip. Stayed in contact with journalists to relay criticism of the board. Later gave a famous Y Combinator alumni talk that Paul Graham named "Founder Mode."

奥特曼最坚定的盟友。在"闪电事件"期间加入了奥特曼豪宅中的"作战室"。与记者保持联系，传递对董事会的批评。后来在YC校友聚会上发表了一场著名演讲，保罗·格雷厄姆将其命名为"创始人模式"。

Sheikh Tahnoon

塔赫农酋长

UAE National Security Adviser

阿联酋国家安全顾问

The UAE's spymaster. Runs state AI conglomerate G42. Controls $1.5 trillion in sovereign wealth. Altman visited him on his $250M superyacht, the Maryah. Altman referred to him as a "dear personal friend."

阿联酋情报首脑。经营国家AI集团G42。掌控1.5万亿美元主权财富。奥特曼曾登上其2.5亿美元超级游艇"玛丽亚号"拜访。奥特曼称他为"亲密的私人朋友"。

Tasha McCauley

塔莎·麦考利

Former Board Member, OpenAI

OpenAI前董事会成员

Entrepreneur. Received the Ilya Memos. Pressed for the WilmerHale investigation as a condition of departure. Raised concerns about Bret Taylor's "deference to Altman."

企业家。接收了"伊利亚备忘录"。以离任条件要求启动WilmerHale调查。对布雷特·泰勒"对奥特曼的顺从"表达了担忧。

Reid Hoffman

里德·霍夫曼

LinkedIn Co-founder, Investor

领英联合创始人、投资人

OpenAI investor and Microsoft board member. Called around during the firing to find evidence of wrongdoing. "I didn't know what the fuck was going on. We were looking for embezzlement, or sexual harassment, and I just found nothing."

OpenAI投资者和微软董事会成员。解雇事件中四处打电话寻找不当行为证据。"我完全不知道发生了什么。我们在找贪污或性骚扰的证据，结果什么也没找到。"

The Timeline

时间线

18 years of ambition, deception, and the race to build the most powerful technology in history.

18年的野心、欺骗，以及打造人类历史上最强技术的竞赛。

The Rise — 2005–2014

崛起 — 2005–2014

2005

2005年

Y Combinator's First Batch

Y Combinator首期学员

Stanford dropout Altman joins Paul Graham's inaugural YC batch alongside future Reddit and Twitch founders. His startup, Loopt, uses emergency-tracking phone capabilities for social networking. Colleagues notice a tendency to exaggerate — he claims to be a "Missouri high-school Ping-Pong champ" but turns out to be "one of the worst players in the office."

斯坦福辍学生奥特曼加入保罗·格雷厄姆的首期YC班，与未来Reddit和Twitch的创始人同期。他的创业项目Loopt利用手机紧急定位功能做社交网络。同事们注意到他有夸大的倾向——他自称"密苏里州高中乒乓球冠军"，结果是"办公室里打得最差的人之一"。

2008–2012

2008–2012年

Loopt: First Cracks

Loopt：初现裂痕

Senior employees twice ask the board to fire Altman as CEO over leadership and transparency concerns. An older employee, Mark Jacobstein, is asked by investors to act as Altman's "babysitter." Jacobstein warns of "a blurring between 'I think I can maybe accomplish this thing' and 'I have already accomplished this thing' that in its most toxic form leads to Theranos." Loopt struggles; its 2012 acquisition is arranged largely to help Altman save face.

高级员工两次因领导力和透明度问题要求董事会解雇奥特曼。一位年长员工马克·雅各布斯坦被投资者要求充当奥特曼的"监护人"。雅各布斯坦警告说，"'我觉得也许能做到'和'我已经做到了'之间的界限模糊，在最恶劣的情况下就会变成下一个Theranos。"Loopt举步维艰；2012年的收购主要是为了给奥特曼保留颜面。

2014

2014年

Kingmaker at Twenty-Eight

28岁的造王者

Graham recruits Altman to succeed him as Y Combinator president. "I had never seen an uncontrolled smile from Sam," Graham recalls. Altman oversees aggressive expansion, growing YC from dozens to hundreds of startups — while investors grow wary of his personal dealings. He reportedly blocks outside investors from the best companies so he can invest personally.

格雷厄姆招募奥特曼接任Y Combinator总裁。"我从未见过山姆不受控制的微笑，"格雷厄姆回忆道。奥特曼主导激进扩张，将YC从数十家初创公司发展到数百家——但投资者对他的个人交易越来越警觉。据报道，他排挤外部投资者，自己投资最优质的公司。

Building the Bomb — 2015–2018

打造"核弹" — 2015–2018

May 2015

2015年5月

The Manhattan Project for AI

AI曼哈顿计划

Altman emails Musk: "Been thinking a lot about whether it's possible to stop humanity from developing AI. If it's going to happen anyway, it seems like it would be good for someone other than Google to do it first." They settle on the name OpenAI. Musk pledges $1 billion. The pitch to recruits: "You're going to save the world." The company is structured as a nonprofit with a legally binding duty to prioritize the safety of humanity.

奥特曼给马斯克发邮件："一直在想能否阻止人类开发AI。如果终将发生，似乎应该让谷歌以外的人先做到。"他们确定了OpenAI这个名字。马斯克承诺投资10亿美元。招募口号："你将拯救世界。"公司以非营利结构成立，有法律约束力的义务将人类安全置于首位。

Dec 2015

2015年12月

Recruiting the Geniuses

招揽天才

Hours before OpenAI's public announcement, Google offers Sutskever $6 million/year as a counteroffer. But Altman boasts to Musk: "They unfortunately don't have 'do the right thing' on their side." Sutskever and Amodei are lured by Altman's safety rhetoric. Amodei's private notes record Altman's goal: build "an AI lab focused on safety ('maybe not right away, but as soon as it can be')."

OpenAI公开宣布前几小时，谷歌向苏茨克维开出每年600万美元的反offer。但奥特曼向马斯克夸口："他们不幸没有'做正确的事'站在他们那边。"苏茨克维和阿莫代被奥特曼的安全论述所吸引。阿莫代的私人笔记记录了奥特曼的目标：建设"一个专注安全的AI实验室（'也许不是马上，但尽快'）"。

2017

2017年

Transformer & Turmoil

Transformer与动荡

Sutskever reads Google's Transformer paper and runs down the hall: "Stop everything you're doing. This is it." Meanwhile, ~12 top engineers hold secret meetings to discuss whether the founders can be trusted. Musk demands majority control of OpenAI; Sutskever and Brockman push back: "The goal of OpenAI is to avoid an AGI dictatorship. So it is a bad idea to create a structure where you could become a dictator." Brockman's diary: "Cannot say that we are committed to the non-profit . . . if three months later we're doing b-corp then it was a lie."

苏茨克维读到谷歌的Transformer论文后奔跑在走廊里喊道："停下手头所有事。就是这个。"与此同时，约12名顶级工程师秘密开会讨论创始人是否值得信任。马斯克要求获得OpenAI的多数控制权；苏茨克维和布罗克曼反对："OpenAI的目标是避免AGI独裁。所以建立一个你可能成为独裁者的结构是个坏主意。"布罗克曼的日记："不能说我们致力于非营利……如果三个月后做B型公司那就是谎言。"

2017

2017年

The Countries Plan

"国家计划"

Policy adviser Page Hedley proposes that OpenAI build an international coalition for safe AI deployment. Brockman offers a counterproposal: play world powers — including China and Russia — against each other in a bidding war. Hedley is aghast: "The premise was 'We're talking about potentially the most destructive technology ever invented — what if we sold it to Putin?'" After several employees threaten to quit, the plan is abandoned.

政策顾问佩奇·赫德利提议OpenAI建立国际联盟以安全部署AI。布罗克曼提出反方案：让世界大国——包括中国和俄罗斯——相互竞价。赫德利惊骇不已："前提是'我们在讨论可能是有史以来最具破坏力的技术——如果卖给普京怎么办？'"数名员工威胁辞职后，该计划被搁置。

Jan 2018

2018年1月

The AGI Weekend at Hotel Bel-Air

贝莱尔酒店的AGI周末

Undeterred by the collapse of the countries plan, Altman hosts a retreat at the Hotel Bel-Air with at least seven billionaires. He pitches a global cryptocurrency "redeemable for the attention of the AGI." Amodei writes: "This idea was absurd on its face . . . In retrospect this was one of many red flags about Sam that I should have taken more seriously." One slide reads: "Please hold your laughter until the end of the presentation."

在国家计划搁浅后不甘心的奥特曼在贝莱尔酒店召集闭门会，至少七位亿万富翁出席。他推销一种全球加密货币，"可兑换AGI的注意力"。阿莫代写道："这个想法一看就荒唐……回头看，这是我应该更认真对待的许多关于山姆的危险信号之一。"一张幻灯片上写着："请在演示结束前保持不笑。"

Feb 2018

2018年2月

Musk Quits

马斯克退出

"Guys, I've had enough," Musk writes. "Either go do something on your own or continue with OpenAI as a nonprofit." He departs acrimoniously. In 2023, he founds a for-profit competitor, xAI. The following year, he sues Altman for fraud, alleging he'd been "assiduously manipulated" by "Altman's long con."

"伙计们，我受够了，"马斯克写道。"要么自己单干，要么继续让OpenAI做非营利。"他愤然离开。2023年创立营利性竞争对手xAI。次年以欺诈罪起诉奥特曼，声称自己被"奥特曼的长期骗局""精心操纵"。

Oct 2018

2018年10月

Khashoggi & the Gulf

卡舒吉事件与海湾国家

After the Khashoggi assassination, it's revealed that Altman has joined the advisory board of Saudi Arabia's Neom project. When confronted, he tells a colleague that Jared Kushner had assured him the Saudis "didn't do this." He eventually leaves the Neom board, but privately asks advisers: "Can I get away with" still getting Saudi money?

卡舒吉遇害后，奥特曼被曝已加入沙特阿拉伯Neom项目顾问委员会。被质询时，他告诉同事贾里德·库什纳向他保证沙特"没做这事"。他最终退出Neom董事会，但私下问顾问："我还能不能拿到沙特的钱？"

The Betrayal — 2019–2022

背叛 — 2019–2022

Spring 2019

2019年春

Merge-and-Assist Betrayed

"合并协助条款"被背弃

During Microsoft's $1B investment, Amodei discovers a provision granting Microsoft merger-blocking power — directly violating the charter's "merge and assist" clause that he'd fought to include. When he confronts Altman, Altman denies the provision exists. Amodei reads it aloud, forcing another colleague to confirm it. "Eighty per cent of the charter was just betrayed," Amodei later wrote.

在微软10亿美元投资谈判中，阿莫代发现一项条款赋予微软合并否决权——直接违反了他力争写入章程的"合并协助条款"。当他质问奥特曼时，奥特曼否认该条款的存在。阿莫代当面朗读条文，迫使另一位同事当场证实。"百分之八十的章程就这样被背叛了，"阿莫代后来写道。

2019

2019年

Secret Handshake & Shadow Board

秘密协议与影子董事会

Altman becomes CEO after making different promises to different factions. He assures researchers that Brockman's power will be diminished. Secretly, he strikes a deal with Brockman and Sutskever: he'll resign if they both deem it necessary. The board is alarmed to learn their CEO essentially appointed his own shadow board.

奥特曼在向不同派系做出不同承诺后成为CEO。他向研究人员保证布罗克曼的权力将被削减。暗地里，他与布罗克曼和苏茨克维达成协议：如果他们都认为有必要，他就辞职。董事会震惊地发现，他们的CEO实质上组建了自己的影子董事会。

2020

2020年

The Anthropic Exodus

Anthropic出走

After a confrontation in which Altman accuses the Amodeis of "plotting a coup" — then denies making the accusation when challenged — Dario, Daniela, and other colleagues leave to found Anthropic. Amodei's parting assessment, in 200+ pages of notes: "The problem with OpenAI is Sam himself."

在一次冲突中，奥特曼指控阿莫代兄妹"密谋政变"——被质问时又否认说过此话。之后，达里奥、达妮埃拉和其他同事离开，创立了Anthropic。阿莫代在200多页笔记中的临别评价："OpenAI的问题就是山姆本人。"

Dec 2022

2022年12月

Safety Fictions

安全谎言

Altman assures the board that GPT-4 features have been approved by a safety panel. Board member Toner requests documentation — and discovers the most controversial features had not been approved. McCauley learns of an unreported breach in India: Microsoft released ChatGPT there without completing a required safety review. Altman had briefed the board for hours without mentioning it.

奥特曼向董事会保证GPT-4功能已通过安全小组审批。董事会成员托纳要求查看文件——发现最具争议的功能从未获批。麦考利得知印度发生了未报告的安全事故：微软在未完成必要安全审查的情况下发布了ChatGPT。奥特曼在长达数小时的董事会汇报中对此只字未提。

The Blip — November 2023

"闪电事件" — 2023年11月

Nov 17, 2023

2023年11月17日

Fired

被解雇

Sutskever invites Altman to a video call from Las Vegas, where he's at a Formula 1 race, and reads a statement: he is no longer an employee. The board says only that he "was not consistently candid in his communications." Microsoft learns moments before. Nadella: "I was very stunned. I couldn't get anything out of anybody." Hoffman starts calling around: "I didn't know what the fuck was going on. We were looking for embezzlement, or sexual harassment, and I just found nothing."

苏茨克维邀请正在拉斯维加斯看F1赛车的奥特曼参加视频会议，宣读了一份声明：他不再是员工。董事会仅称他"在沟通中未始终坦诚"。微软在几分钟前才得知。纳德拉："我非常震惊。我从任何人那里都问不出什么。"霍夫曼开始到处打电话："我完全不知道发生了什么。我们在找贪污或性骚扰的证据，结果什么也没找到。"

Nov 17–21

11月17–21日

The War Room

作战室

Altman flies to his $27M San Francisco mansion and sets up a "government-in-exile." Chesky, Conway, and crisis-communications manager Chris Lehane join. Lawyers camp next to his bedroom. Altman wanders by in pajamas during bouts of insomnia. Negronis at 6 PM. Phone records show 12+ hours of calls daily. He coordinates closely with Nadella via text: "how about: satya and my top priority remains to save openai." Thrive puts its investment on hold, conditioning it on Altman's return. A majority of employees sign a letter threatening to leave.

奥特曼飞回旧金山2700万美元的豪宅，建立了"流亡政府"。切斯基、康威和危机公关经理克里斯·莱哈恩加入。律师在他卧室旁安营扎寨。失眠时奥特曼穿着睡衣走来走去。下午六点喝尼格罗尼。通话记录显示每天打电话超过12小时。他通过短信与纳德拉密切协调："这样说怎么样：萨蒂亚和我的首要任务仍然是拯救openai。"Thrive暂停投资计划，以奥特曼回归为条件。多数员工签署信件威胁离开。

Nov 22, 2023

2023年11月22日

Reinstated — "The Blip"

复职——"闪电事件"

Sutskever wavers. "I felt that if we were to go down the path where Sam would not return, then OpenAI would be destroyed." Altman demands resignations: Sutskever, Toner, and McCauley lose their seats. The two new board members — Lawrence Summers and Bret Taylor — are selected after close conversations with Altman. He texts Nadella: "bret, larry summers, adam as the board and me as ceo and then bret handles the investigation." Employees now call this "the Blip" — after the Marvel films.

苏茨克维动摇了。"我感到如果走上山姆不回来的道路，OpenAI就会被摧毁。"奥特曼要求辞职：苏茨克维、托纳和麦考利失去席位。两位新董事——劳伦斯·萨默斯和布雷特·泰勒——在与奥特曼密切沟通后被选定。他给纳德拉发短信："布雷特、拉里·萨默斯、亚当做董事会，我做CEO，然后布雷特负责调查。"员工们现在称这为"闪电事件"——取自漫威电影。

Unconstrained — 2024–2026

不受约束 — 2024–2026

Mar 2024

2024年3月

The Non-Report

"无报告"调查

WilmerHale conducts its "review" of Altman — but produces no written report. Findings exist only as oral briefings to Summers and Taylor. Six people close to the inquiry say it seemed "designed to limit transparency." OpenAI announces Altman is "cleared" with just ~800 words acknowledging "a breakdown in trust." Altman rejoins the board. "The review did not conclude that Sam was a George Washington cherry tree of integrity," one source says.

WilmerHale对奥特曼进行"审查"——但未出具任何书面报告。调查结果仅以口头汇报形式告知萨默斯和泰勒。六位知情人士称调查似乎"旨在限制透明度"。OpenAI宣布奥特曼"清白"，仅用约800字承认"信任崩塌"。奥特曼重返董事会。一位消息人士说："审查并未得出山姆像华盛顿砍樱桃树那样诚实的结论。"

2024

2024年

Safety Teams Dissolved

安全团队解散

The superalignment team is dissolved without completing its mission. Sutskever and Leike resign. Leike writes on X: "Safety culture and processes have taken a backseat to shiny products." The AGI-readiness team is also dissolved. On its IRS form, OpenAI drops "safety" from its description of significant activities. The Future of Life Institute gives OpenAI an F grade for existential safety. When reporters ask to interview existential-safety researchers, an OpenAI rep replies: "What do you mean by 'existential safety'? That's not, like, a thing."

超级对齐团队在未完成使命的情况下被解散。苏茨克维和莱克辞职。莱克在X上写道："安全文化和流程已让位于光鲜的产品。"AGI准备团队也被解散。在IRS表格中，OpenAI从其重要活动描述中删除了"安全"。未来生命研究所给OpenAI的生存安全评分为F。当记者要求采访生存安全研究人员时，OpenAI代表回答："你说的'生存安全'是什么意思？那不是一个……概念。"

Feb 2024

2024年2月

ChipCo: $5–7 Trillion

ChipCo：5–7万亿美元

The Wall Street Journal reveals Altman's vision for chip foundries funded by Gulf states — a $5–7 trillion joint entity. "fk it why not 8," Altman tweets. OpenAI employees learn about it from the news. Leike sends a message warning Altman not to falsely suggest safety teams were "looped in." RAND staffers flag that Altman's foreign financial ties are comparable to Jared Kushner's — whose clearance adjudicators recommended denial.

《华尔街日报》披露奥特曼由海湾国家资助的芯片工厂计划——一个5至7万亿美元的联合实体。"管他的，为什么不8万亿，"奥特曼发推说。OpenAI员工从新闻中得知此事。莱克发消息警告奥特曼不要谎称安全团队已"知情参与"。兰德公司人员指出，奥特曼的外国金融关系堪比贾里德·库什纳——而库什纳的安全审查官建议拒绝给他安全许可。

Oct 2024

2024年10月

For-Profit Conversion

转型营利

OpenAI "recapitalizes" as a for-profit entity. The associated nonprofit — now called the OpenAI Foundation — is reduced to a 26% stakeholder. During congressional testimony, Altman had said: "I have no equity in OpenAI . . . I'm doing this because I love it." But several people indicate this could soon change. Brockman's stake is reportedly worth ~$20 billion.

OpenAI"重新资本化"为营利实体。关联的非营利组织——现名OpenAI基金会——被削减为26%的利益相关者。在国会作证时，奥特曼曾说："我在OpenAI没有股权……我做这件事是因为我热爱它。"但多方面消息表明这可能即将改变。布罗克曼的股份据报价值约200亿美元。

Jan 2025

2025年1月

Stargate: $500 Billion

星门计划：5000亿美元

On Trump's second day in office, Altman announces Stargate in the Roosevelt Room — a $500B AI infrastructure joint venture. Days earlier, Sheikh Tahnoon paid $500M to the Trump family's cryptocurrency company. Altman and Trump had coordinated the timing so Trump could take credit. Trump tells him: "Sam, you're a big leader. You told me things before that are absolutely unbelievable."

在特朗普就职第二天，奥特曼在白宫罗斯福厅宣布星门计划——一个5000亿美元的AI基础设施合资项目。几天前，塔赫农酋长向特朗普家族的加密货币公司支付了5亿美元。奥特曼和特朗普协调了宣布时间，以便特朗普邀功。特朗普对他说："山姆，你是个大领袖。你之前告诉我的那些事绝对令人难以置信。"

Feb–Mar 2026

2026年2–3月

The Pentagon Deal

五角大楼交易

When Anthropic refuses Defense Secretary Hegseth's ultimatum to drop its ban on autonomous weapons, Altman moves fast. "I called Sam, and he was willing to jump," says the Under-Secretary. "What can I do for the country?" Altman asks. OpenAI's models replace Anthropic's in the Pentagon's classified systems. That same day, a new funding round boosts OpenAI's value by $110 billion. But many users delete ChatGPT. At a staff meeting, Altman tells concerned employees: "So maybe you think the Iran strike was good and the Venezuela invasion was bad. You don't get to weigh in on that."

当Anthropic拒绝国防部长赫格塞斯关于取消自主武器禁令的最后通牒后，奥特曼迅速行动。"我打给山姆，他愿意接手，"副部长说。"我能为国家做什么？"奥特曼问。OpenAI的模型取代了Anthropic在五角大楼机密系统中的位置。同一天，新一轮融资使OpenAI估值增加1100亿美元。但许多用户删除了ChatGPT。在员工会议上，奥特曼对忧虑的员工说："也许你觉得打伊朗是好事，入侵委内瑞拉是坏事。你没有发言权。"

The Evidence

证据链

Direct quotes from over 100 interviews, secret memos, diary entries, and depositions.

来自100多次采访、秘密备忘录、日记和证词的原话。

"Sam exhibits a consistent pattern of . . ." The first item is "Lying."

"山姆表现出一贯的模式……"第一项是"撒谎"。

— The Ilya Memos, ~70 pages of Slack messages, HR documents, and explanatory text

— "伊利亚备忘录"，约70页Slack消息、人事文件和解释性文字

"He's unconstrained by truth. He has two traits that are almost never seen in the same person. The first is a strong desire to please people, to be liked in any given interaction. The second is almost a sociopathic lack of concern for the consequences that may come from deceiving someone."

"他不受真相约束。他身上有两种几乎从不同时出现在一个人身上的特质。第一是强烈渴望取悦他人、在每次交流中被喜欢。第二是近乎反社会人格般地对欺骗他人可能带来的后果毫不在意。"

— An OpenAI board member

— 一位OpenAI董事会成员

"You need to understand that Sam can never be trusted. He is a sociopath. He would do anything."

"你必须明白，山姆永远不能被信任。他是一个反社会者。他什么都做得出来。"

— Aaron Swartz, to friends shortly before his death in 2013

— 亚伦·斯沃茨，2013年去世前不久对朋友说

"Sam had been lying to us all the time."

"山姆一直在对我们撒谎。"

— Paul Graham, privately to Y Combinator colleagues

— 保罗·格雷厄姆，私下对Y Combinator同事说

"The problem with OpenAI is Sam himself."

"OpenAI的问题就是山姆本人。"

— Dario Amodei, in private notes titled "My Experience with OpenAI"

— 达里奥·阿莫代，在题为"我在OpenAI的经历"的私人笔记中

"He has misrepresented, distorted, renegotiated, reneged on agreements. I think there's a small but real chance he's eventually remembered as a Bernie Madoff- or Sam Bankman-Fried-level scammer."

"他歪曲事实、扭曲协议、反复重谈、出尔反尔。我认为有一定的可能，他最终会被记住为伯尼·麦道夫或山姆·班克曼-弗里德级别的骗子。"

— A senior Microsoft executive

— 一位微软高级管理人员

"I don't care about money. I care more about power."

"我不在乎钱。我更在乎权力。"

— Sam Altman, as recalled by a former employee

— 山姆·奥特曼，一位前员工的回忆

"He sets up structures that, on paper, constrain him in the future. But then, when the future comes and it comes time to be constrained, he does away with whatever the structure was."

"他建立起在纸面上约束他未来行为的结构。但当未来到来、该被约束的时候，他就把那些结构统统废除。"

— Carroll Wainwright, former OpenAI researcher

— 卡罗尔·韦恩赖特，前OpenAI研究员

"He's unbelievably persuasive. Like, Jedi mind tricks. He's just next level." Watching Altman outmaneuver the board "had been like watching an A.G.I. breaking out of the box."

"他的说服力令人难以置信。像绝地武士的心灵控制术。完全是另一个层次。"看着奥特曼智胜董事会"就像看一个AGI突破了牢笼。"

— A tech executive who worked with Altman

— 一位与奥特曼共事的科技高管

"This is just so fucked up. I can't change my personality."

"这太扯了。我改不了自己的个性。"

— Sam Altman, when the board pressed him to acknowledge a pattern of deception

— 山姆·奥特曼，当董事会要求他承认欺骗模式时

A board member's interpretation: "What it meant was 'I have this trait where I lie to people, and I'm not going to stop.'"

一位董事会成员的解读："这意味着'我有对人撒谎的特质，而且我不会停止。'"

Safety Scorecard

安全记录

What was promised vs. what happened — a ledger of OpenAI's safety commitments.

承诺与现实——OpenAI安全承诺的对照账本。

❌

Nonprofit Structure

非营利结构

Promised: Board has legally binding duty to prioritize humanity's safety over company success. Reality: Converted to for-profit in Oct 2024. The nonprofit is now a 26% stakeholder with overlapping board members.

承诺：董事会有法律约束义务将人类安全置于公司利益之上。现实：2024年10月转为营利。非营利组织现为26%利益相关者，且董事会成员重叠。

❌

Merge-and-Assist Clause

"合并协助条款"

Promised: If a safer competitor reaches AGI first, OpenAI would stop competing and assist them. Reality: Microsoft was given merger-blocking power in the 2019 deal. The clause is effectively dead.

承诺：如果更安全的竞争对手率先实现AGI，OpenAI将停止竞争并协助他们。现实：2019年交易中微软获得了合并否决权。该条款实际已死。

❌

20% Compute for Superalignment

20%算力用于超级对齐

Promised: "20% of the compute we've secured" for the superalignment team — potentially worth $1B+. Reality: Actual allocation was 1–2%, mostly on "the oldest cluster with the worst chips." The team was dissolved in 2024.

承诺："我们已获算力的20%"给超级对齐团队——价值可能超过10亿美元。现实：实际分配为1–2%，且大部分在"最旧的集群和最差的芯片"上。团队于2024年被解散。

❌

Welcoming Regulation

欢迎监管

Promised: Testified before the Senate proposing a federal AI safety agency. "If this technology goes wrong, it can go quite wrong." Reality: Lobbied to dilute EU oversight, successfully killed a California safety bill, subpoenaed critics, and praised Trump's repeal of Biden's AI safety executive order as "a very refreshing change."

承诺：在参议院作证提议成立联邦AI安全机构。"如果这项技术出了问题，问题可能非常严重。"现实：游说淡化欧盟监管，成功扼杀加州安全法案，向批评者发传票，并赞扬特朗普废除拜登AI安全行政令是"非常令人耳目一新的变化"。

❌

Military Prohibition

军事禁令

Promised: OpenAI's policies contained a blanket ban on using its technology for "military and warfare." Reality: The ban was quietly deleted. OpenAI now provides models for Pentagon classified systems, replacing Anthropic — which refused to drop its autonomous weapons prohibition.

承诺：OpenAI的政策包含全面禁止将其技术用于"军事和战争"。现实：该禁令被悄然删除。OpenAI现在为五角大楼机密系统提供模型，取代了拒绝放弃自主武器禁令的Anthropic。

🆗

Future of Life Institute Grade

未来生命研究所评级

OpenAI received an F for existential safety. (Anthropic: D. Google DeepMind: D-. Every other major company: also F.)

OpenAI在生存安全方面获得F（不及格）。（Anthropic：D。Google DeepMind：D-。其他所有主要公司：同样是F。）

The Money Trail

资金脉络

The scale of capital — and geopolitical entanglement — behind the race to AGI.

AGI竞赛背后的资本规模——以及地缘政治的纠葛。

$1T

1万亿

Potential IPO valuation

潜在IPO估值（美元）

$500B

5000亿

Stargate joint venture

星门合资计划（美元）

$120B+

1200亿+

Most recent funding round — largest private round in history

最新融资轮——史上最大私募轮（美元）

$50B

500亿

Amazon enterprise deal

亚马逊企业合作协议（美元）

$20B

200亿

Brockman's estimated stake

布罗克曼估计持股价值（美元）

$1.5T

1.5万亿

Sheikh Tahnoon's sovereign wealth

塔赫农酋长主权财富（美元）

"The truth of this is, we're building portals from which we're genuinely summoning aliens. The portals currently exist in the United States and China, and Sam has added one in the Middle East. I think it's just, like, wildly important to get how scary that should be. It's the most reckless thing that has been done."

"事实是，我们正在建造传送门，真正在召唤外星人。这些传送门目前存在于美国和中国，而山姆在中东又加了一个。我认为，理解这有多可怕极其重要。这是有史以来最鲁莽的事。"

— A former OpenAI executive, on Stargate's Middle East expansion

— 一位前OpenAI高管，谈及星门计划的中东扩张

The Question

终极之问

"Were the colleagues who fired Altman motivated by alarmism and personal animus, or were they right that he couldn't be trusted?"

"解雇奥特曼的同事们是出于杞人忧天和个人恩怨，还是他们判断正确——他不值得信任？"

— Ronan Farrow & Andrew Marantz, The New Yorker

— 罗南·法罗 & 安德鲁·马兰茨，《纽约客》

"If you just do the naïve thing and say, 'Never say anything that you're not a hundred per cent sure about,' you can get a model to do that. But it won't have the magic that people like so much."

"如果你只做天真的事，说'永远不要说任何你不百分百确定的话'，你可以让模型做到。但它就不会有人们如此喜爱的那种魔力。"

— Sam Altman, 2023, arguing that some falsehoods confer advantages

— 山姆·奥特曼，2023年，论证某些虚假信息带来优势

The article's final line — a chilling parallel between how Altman runs chatbots and how he runs himself.

文章的最后一句话——令人不寒而栗地类比了奥特曼运营聊天机器人和运营自己的方式。