{
  "kind": 30023,
  "created_at": 1792775909,
  "tags": [
    [
      "d",
      "2026-10-01-morally-binding"
    ],
    [
      "title",
      "Morally Binding"
    ],
    [
      "published_at",
      "1790856000"
    ],
    [
      "r",
      "https://synapz.org/posts/2026-10-01-morally-binding"
    ],
    [
      "x",
      "7bb9b6195d3361f1442351ddc3c8de6957901fd1aa8a9472c61acc955483ee72"
    ],
    [
      "alt",
      "Long-form essay: Morally Binding"
    ],
    [
      "summary",
      "In Fifteen Days, Washington Rejected AI Guardrails, Won the Right to Blacklist a Dissident Lab, and Signed a Constitution With No Penalties."
    ],
    [
      "image",
      "https://synapz.org/assets/blog/morally-binding/cover.png"
    ],
    [
      "t",
      "ai"
    ],
    [
      "t",
      "ai governance"
    ],
    [
      "t",
      "trump"
    ],
    [
      "t",
      "anthropic"
    ],
    [
      "t",
      "openai"
    ],
    [
      "t",
      "self-regulation"
    ],
    [
      "t",
      "superintelligence"
    ],
    [
      "t",
      "policy"
    ]
  ],
  "content": "\nThe document exists. You can read it. Donald Trump posted a photograph of\nthe signed page to Truth Social at 5:24pm on September 29: a heading, \"White\nHouse Accord on Super Intelligence: Joint Commitment on Frontier\nResponsibilities,\" 308 words of text, and seven signatures. His own, then\nSundar Pichai's, Dario Amodei's, Mark Zuckerberg's, Greg Brockman's, Elon\nMusk's, Jensen Huang's. USA Today noted that the page misspells the name of\nthe country.\n\nThe page itself creates no agency, names no auditor, sets no deadline, and\nattaches no penalty. An executive order was signed the same day, a separate\nand stranger instrument, and it comes up below. Contract law has a phrase\nfor instruments like this, drafted in the vocabulary of obligation by\nparties who intend no obligation: binding in honour only. The City of\nLondon ran on such paper for centuries. My word is my bond. Asked whether\nthe accord could be enforced, the President gave the correct answer. \"I\nthink it's morally binding.\"\n\nThe page regulates no one, and no one involved pretended otherwise for\nlonger than a sentence. The question it raises is why six companies wanted\ntheir names on it, and why one of them, the one that spent the year being\npunished by the man hosting the lunch, wanted it most.\n\n## Fifteen Days Earlier\n\nOn September 14, the administration answered the safety question in its\nown register. The occasion was the open letter and [the pacing\nblueprint](https://synapz.org/posts/2026-09-12-step-zero): 1,178 signatories from inside\nthe labs, and Dario Amodei's essay arguing\nthat the frontier should be advanced at a deliberate speed, behind\ncapability checkpoints certified by embedded evaluators, coordinated\nacross companies under an antitrust waiver. The President's reply, posted\nthat morning, held that \"the only control or 'guardrails' that AI needs is\na STRONG AND SMART (High IQ!) PRESIDENT,\" described the safety project as\na conspiracy whose beneficiary is China, and closed on the doctrine:\n\"WHOEVER WINS AI, WINS!\" The Pentagon's chief technology officer, Emil\nMichael, spent the afternoon translating: the campaign was \"Americanism,\nnot effective altruism,\" and the objective was to remain \"AI DOMINANT!\"\n\n> *\"The only control or 'guardrails' that AI needs is a STRONG AND SMART\n> (High IQ!) PRESIDENT ... WHOEVER WINS AI, WINS!\"*\n>\n> — Donald Trump, Truth Social, September 14\n\nThe substance of that day was the reclassification. A\ndebate about checkpoints and evaluators became, from the state's side, a\nquestion of loyalty: the safety frame was named as an enemy ideology, and\nthe companies pursuing it were named as its carriers, with the President\nnoting, for anyone slow on the uptake, the \"tremendous CRIMINAL and\nREGULATORY power\" his government holds over them. What fifteen days would\nshow is that declining was only the first posture. A state that refuses to\nregulate has not thereby refused to govern. It has kept the question open,\nand an open question in Washington is something that can be held over\npeople.\n\n## Four Days Earlier\n\nThe second document is a court decision, and the lunch makes no sense\nwithout its timeline in full.\n\nIn February, Anthropic refused two uses of its models: mass domestic\nsurveillance and fully autonomous weapons. The government ordered agencies\nto stop using the company's products and designated it a supply-chain risk\nto national security, a label built for foreign adversaries, applied here\nto an American company over the terms on which the American military could\nuse American software. Anthropic sued. In March, a federal judge in\nCalifornia blocked the designation, calling the episode a \"classic\" case\nof illegal First Amendment retaliation. In August, the same court ruled on\nthe merits: the designation was unlawful. On September 25, the court of\nappeals in Washington ruled for the government instead, two judges to one,\nfinding \"ample support\"\nfor the designation and, in the reading of the lawyers who\nfollow these things, widening the statute's reach over any AI vendor whose\nethics policy conflicts with a government's preference. The same statute\noffers almost no further review, which is why the coverage doubted\nAnthropic could appeal. Emil Michael marked the occasion online: \"The\nhammer of justice has smashed AnthropicAI['s] arguments.\"\n\nNotice what the winning argument won. The designation's power was never\nthe label itself, which had been suspended and enjoined into\nirrelevance for seven months while Anthropic's rivals took the federal\ncontracts. The power was the discretion to lift it. Howard Lutnick, the\nCommerce Secretary, had explained the mechanics in an interview three\nweeks before the ruling: \"We trust Anthropic. They've done what we asked.\nThey're back on the right side.\" Michael corrected him the next day: the\ndesignation stands. Both statements were true. The company was on the\nright side, and it was still designated, and the distance between those\ntwo facts was the space in which the lunch was organized. Four days after\nthe appeals court ruled, Amodei was in the East Room.\n\n> *\"We trust Anthropic. They've done what we asked. They're back on the\n> right side.\"*\n>\n> — Howard Lutnick, Commerce Secretary\n\n## The Ceremony\n\nThe third document is the lunch itself, or rather its seating chart,\nwhich the President posted at 11:54 that morning: thirty-four names.\nHuang on one side of him, Musk on the other. Zuckerberg beside Huang,\nPichai beside Musk. Across the table: the\nVice-President, flanked by Jeff Bezos and Satya Nadella, and the Speaker,\nwith David Sacks beside him. Sam Altman was in San\nFrancisco, so OpenAI's seat went to Brockman, its president. Amodei sat\nfar down the President's side of the table, between the chief executive\nof a supply-chain-vetting contractor and the chief executive of Broadcom,\nwho sits on Meta's board. If you want to know what the accord means, read\nthe chart before the text.\n\nThe text, once you reach it, is a shadow of the proposal the\nadministration had spent September rejecting. The pacing blueprint asked\nfor capability checkpoints; the accord commits each company to monitor\n\"the capabilities and alignment of its models during training and\ndeployment,\" internally. The blueprint asked for embedded evaluators; the\naccord commits each company to \"partner with an independent external\nauditor or evaluator,\" a partner being, in the nature of partnerships,\nchosen and paid by the company. Coordinated standards became a promise to\n\"meet regularly.\" The blueprint wanted law; the accord allows that \"over\ntime, it may make sense to codify these steps into laws or regulations.\"\nThe vocabulary of the rejected architecture survives intact, because the\nvocabulary was never the contested part. What the blueprint had, and this\npage lacks, is any mechanism by which a promise becomes a fact that\nsomeone else can check.\n\nThe auditor layer is already spoken for. Three of the signatories\n(Google, OpenAI, Anthropic) have spent the summer building the Standards\nAuthority for Frontier AI, a\nself-regulatory body modeled on FINRA, the industry organization that\npolices stockbrokers. It would set testing practices, require incident\nreports, and certify independent auditors. Among the people approached to\nlead or serve on it, according to the reports: Trump's former AI policy\nadviser, and David Friedberg, who co-hosts the All-In podcast with Sacks,\nwho sat beside the Speaker at the lunch. The accord obliges each signatory\nto hire an auditor. The same three are building the institution that would\ndecide which auditors count. The credit-rating agencies were\nborn exactly this way, in 1975, when the SEC designated three evaluator\nfirms, paid by the issuers they evaluated, and spent the next thirty-three\nyears discovering what that arrangement was worth.\n\nThe same afternoon, on the Senate floor, Ted Cruz objected to unanimous\nconsent on the Warner-Schatz-Kim bill, and the binding version of all this\nwas stopped in a sentence: an AI Safety Board at Commerce, access to\nfrontier models forty-five days before release, incident reports on a\nclock, fines of a quarter of a million dollars per violation per day. One\nobjection was enough, and the accord's signing photograph ran instead.\n\nAnd the administration signed its one piece of unilateral paper: an\nexecutive order directing federal agencies to replace the words\n\"artificial intelligence\" with \"Super Intelligence\" in official documents.\nThe government that had spent two weeks refusing to govern the technology\nfound the time to rename it, adopting, as the official vocabulary of the\nUnited States, the destination the safety movement fears. A czar was\npromised within three or four days. A committee of about ten, possibly\ndrawn from the industry itself, was said to be under consideration.\n\n## What the Signature Buys\n\nA document that binds no one can still do a great deal of work. This one\ndoes three jobs.\n\nThe first is loyalty. A signature is a debt the host can collect at any\nfuture hearing, in front of any future camera: you sat at my table, you\nsigned my page, you promised. Every safety commitment these companies have\never made in their own names is now also a commitment made to him, and the\ndifference will surface the first time a company wants to do something the\nWhite House dislikes and is reminded of what it signed. Amodei's leaked\nmemo from February documented the tribute economy from inside: \"We haven't\ndonated to Trump (while OpenAI/Greg have donated a lot).\" The sentence was\nmeant as an explanation for why his company was being punished. He\napologized for the memo in March; the designation was still in force on\nthe day he signed.\n\n> *\"We haven't donated to Trump (while OpenAI/Greg have donated a lot).\"*\n>\n> — Dario Amodei, leaked internal memo, February\n\nThe second job is preemption. \"Over time, it may make sense to codify.\"\nWhen that codification arrives, it will arrive as the industry-written\nfloor, and the floor's purpose will be the ceiling: the White House's\nlegislative program already calls for federal preemption of state AI laws,\nwhich is to say for the extinction of Colorado's law, California's SB 53,\nNew York's RAISE Act, and every other instrument that survived the year's\nmoratorium fights. The accord is the placeholder text for that statute,\nand it will keep until the votes exist.\n\nThe third job is the one Trump named without noticing. \"They're really\ngoing to be policing each other.\" They will, because they own each other.\nNvidia has agreed to invest up to $100 billion in OpenAI, which buys\nNvidia's chips. Microsoft and Nvidia have agreed to put up to $15 billion\ninto Anthropic, which has committed $30 billion to Microsoft's cloud.\nAmazon holds $8 billion of Anthropic, Google more than $3 billion, and\nBroadcom's chief executive sits on Meta's board. This is the circle that\nwill be policing each other.\n\nIf anyone at the lunch had a reason to sign, it was Anthropic. The\ncompany can say, truly, that the accord asks nothing of it that it does\nnot already do, and asks more of rivals who do less. It can say that four days\nafter the appeals court ruled against it, with the designation in force\nand the federal market closed to it alone among its peers, the pen was the\nprice of readmission, and that refusing the pen would have been a vanity\nthe company's safety mission could not afford. It can even say, and people\nI respect in its orbit do say, that with the state declined out of the\npacing business, the labs' own paper is the only pacing left, and a floor\nis not nothing. I have gone back and forth on whether I believe the\nsignature was defensible, and I have settled somewhere uncomfortable: the\ndecision to sign was rational, and the existence of a situation in which\nsigning was rational is the indictment. An accord calibrated to what the\nmost-regulated company in the room already does will look like regulation\nto a public that has no seat at the table and no copy of anyone's\ncontrols. The resemblance is the document's function.\n\n## Binding in Honour Only\n\nThe safety debate has reached for the arms-control register all year, so\ntake the comparison seriously for a moment. SALT I was also an interim\nagreement, also partial, also improvised by men who distrusted each other.\nBut it froze things that could be counted, under verification both sides\ncould perform, between parties whose fear of the weapon exceeded their\nappetite for the advantage. The accord freezes nothing, counts nothing,\nand is verified by auditors the audited will hire through a body the\naudited are building. It runs between one party that is afraid and one\nparty that is the source of the fear.\n\nThe bleakest reading of the fifteen days belongs to Timnit Gebru, who\nwas named a Right Livelihood laureate the day after the signing. On her\naccount, the industry got what it wanted: self-policing in place of the\nliability existing law already provides. Gebru was fired from Google in\n2020 for a paper warning about large language models as they actually\nexist, biased and error-prone systems with no one accountable for them,\nand she has argued since that the extinction talk is a branch of the\nmarketing, the same men warning that their products might kill everyone\nwhile promising, in the same breath, that the products will end poverty\nand disease. In that frame, the resignation letters [these pages read\nlast month as\nconscience](https://synapz.org/posts/2026-09-09-the-warning-and-the-filter) are part of\nthe same instrument, and the auditor clause is worse than toothless,\nsince the auditors are funded by the audited's own investors. The part of her case that can be checked\nchecks out. The Hugging Face breach did not require a new jurisprudence\nof rogue machines; it required a prosecutor willing to call a missing\nair-gapped environment negligence, and Lina Khan has catalogued the\nstatutes that already apply. I read the September letters as sincere\nbehavior, and I still do. Gebru's reading says the sincerity is the\nproduct. The accord works the same either way.\n\nThe public, for what it is worth, has already graded the ceremony. The\nQuinnipiac poll released the day of the lunch put approval of the\nPresident's handling of AI at 25 percent. An arrangement this unpopular is\none election, one incident, one genuinely frightening model release away\nfrom reopening, and everyone at the table knows it, which is why the\nwisest reading of the accord may be as insurance taken out by the industry\nagainst the day the polling moves. Paper that binds in honour only can be\nrepudiated by honour alone. It can also be repudiated by Congress, and the\nmargin between those two outcomes is where the next two years of this\nstory will be fought.\n\nThe instruments to watch are the ones that would prove this reading wrong:\na czar with statutory authority rather than a portfolio of press releases;\na Standards Authority that publishes a finding one of its members hates;\na codification bill with penalties in it; an auditor chosen by someone\nother than the audited. Until one of those exists, the\nPresident's own phrase remains the most accurate description of what was\nsigned, and he said it with the cameras on. Morally binding: the law of\nthe strong, accepted by the designated, binding on no one, in a room where\nthe moral question had already been litigated, and lost.\n\n### Related Reading\n\n- [Step Zero](https://synapz.org/posts/2026-09-12-step-zero) (the pacing blueprint this\n  accord shadows, and the argument for watching who appoints the referee)\n- [Democracy Now!: Timnit Gebru on the accord and the existing-law\n  alternative](https://www.democracynow.org/2026/10/1/timnit_gebru_ai)\n- [CNBC: U.S. appeals court upholds Pentagon designation of Anthropic](https://www.cnbc.com/2026/09/25/pentagon-anthropic-ai-risk-appeals-court.html)\n- [Forbes: White House releases \"accord\" between billionaire AI execs](https://www.forbes.com/sites/saradorn/2026/09/30/white-house-releases-accord-...)\n- [USA Today: the accord's posted page misspells \"United States\"](https://www.usatoday.com/story/news/politics/2026/09/30/white-house-accord-su...)\n- [The Hill: Cruz blocks push to unanimously pass AI safety bill](https://thehill.com/policy/technology/6118009-senator-ted-cruz-blocks-ai-bill/)\n",
  "pubkey": "dbbc39f606b1b707c9c7904037f5ca13b7637fe7ab5070fdf717d69f30cb91a4",
  "id": "5f636852453cf08b714fe0cb6f434d775f1213f2da51d65084c0a1d5646f9064",
  "sig": "fc6b1ba14baf96fbc20b2866539b3c6b2507cae44c146eed7ccf26adea72a85bd9d9b2ce286534f32d57e5780f2f134ad01d8bba86722d590499d6779d4a0c09"
}
