White House Accord on Super Intelligence  🔗 **Source:** [White House Accord on Super Intelligence](https://twitter.com/WhiteHouse/status/2105292669303791687/?rw_tt_thread=True) > [!note]+ Document Notes > tags: [ai-governance, frontier-ai-safety, voluntary-commitments, third-party-audits, ai-policy] > author: The White House > source: X (Twitter), @WhiteHouse > reference: https://twitter.com/WhiteHouse/status/2105292669303791687 > date created: 2026-10-04 > > > [!info] Generated analysis — not the author's words. Sections below are machine-written > > from the captured source. Only text inside `> [!quote]` blocks is verbatim from the author. > > Do not quote or attribute any other line to the named author. > <!-- deep-read:sections start --> ## 🏷️ Title **"White House Accord on Super Intelligence"** — posted by @WhiteHouse on X, September 30, 2026. The post text is the six-word title. Everything substantive is in two attached images: a one-page "Joint Commitment on Frontier Responsibilities" and a signature page. **What the images contain (transcribed for this analysis).** Page 1 is about 330 words: a preamble, four "layers of controls and audits," a commitment to "meet regularly to establish standards and best practices," and a closing line that "it may make sense to codify these steps into laws or regulations." Page 2 carries seven signatures: **Donald J. Trump**, "President of the Unites States" (sic), and six company heads — **Sundar Pichai** (Google), **Dario Amodei** (Anthropic), **Mark Zuckerberg** (Meta), **Greg Brockman** (OpenAI), **Elon Musk** ("XAI"), and **Jensen Huang** (Nvidia). **Disclosure.** This analysis is produced by a Claude model, built by Anthropic, one of the six signatories. Read the assessment of Anthropic's commitment with that in mind. **Verification status.** The doc came from the official @WhiteHouse account per Reader's capture. This analysis didn't find a whitehouse.gov text version or a press release; the accord's legal status (executive order, memorandum, or press artifact) is unverified. ## 🧭 Table of Contents 1. [References (in-page)](#-references-in-page) 2. [Summary](#-summary) 3. [Main Content](#-main-content) → [Notes](#️-notes) 4. [References](#-references) 5. [Paraphrase](#️-paraphrase) 6. [Newsletter](#-newsletter) 7. [Action Items](#-action-items) 8. [Hypotheses](#-hypotheses) 9. [Methodology](#-methodology) 10. [Questions](#-questions) 11. [Future Research](#-future-research) 12. [Implications](#-implications) 13. [Learning Styles Assessment](#-learning-styles-assessment) 14. [Accessibility Assessment](#-accessibility-assessment) 15. [DEIB Assessment](#-deib-assessment) 16. [Logical Validity](#-logical-validity) 17. [DARVO Analysis](#️-darvo-analysis) 18. [Conclusion](#-conclusion) 19. [Refactor Appendix](#-refactor-appendix) ## 🔗 References (in-page) The markup was enumerated before any sourcing claim was made. **No links.** The captured HTML is one title paragraph and two `<img>` elements: - `https://pbs.twimg.com/media/HTeAeipWoAEZ3vF.jpg` — page 1, the commitment text. `alt="Image"`. - `https://pbs.twimg.com/media/HTeAZHCW8AAUDdP.jpg` — page 2, signatures. `alt="Image"`. No link to a whitehouse.gov page, a PDF, or any company statement. The absence is a real finding: the full public record of this accord, as captured, is two images of a scanned document. ## 📝 Summary - **What happened.** The White House posted a one-page voluntary accord in which six frontier AI and chip companies commit to four layers of internal controls and audits for models "during training and deployment." - **Why it matters.** It's the first signed, multi-company frontier-safety commitment from this administration that names independent external audit and board-level oversight. Its title says "Super Intelligence"; its body never uses the term. - **Key stakeholders.** The six signatory companies; companies not on the list (Microsoft, Amazon, Mistral, DeepSeek and other non-US labs are absent); third-party evaluators (who'd do the audits); Congress (codification is floated); the public. - **Risks and tradeoffs.** Everything is voluntary and self-defined: "robust," "independent," and "operating as intended" have no stated criteria. No timeline, no reporting requirement, no enforcement. The text is published only as images. - **Bottom line.** A short, plausible governance skeleton — internal controls, an internal team, an external auditor, a board committee — with no specifics, no accountability mechanism, and an inaccessible format. ## 📖 Main Content **Preamble.** "Every company is responsible for developing its own technology safely and in a way that builds trust with customers and the public." The framing puts responsibility on each company, not on government. **The four layers.** 1. **Internal controls** to "monitor the capabilities and alignment of its models during training and deployment around areas like cybersecurity, biosecurity, and chemical threats," and to ensure models "do not hack or access technical systems in unintended ways." 2. **An internal team** empowered to ensure controls, monitoring and detection "are operating as intended," and that issues are remediated. 3. **An independent external auditor or evaluator** to assess whether the controls work. 4. **An independent board committee** to oversee and receive reports from the internal team and the external auditors, and to ensure issues are remediated. **Process commitment.** "The participating companies will meet regularly to establish standards and best practices." **Codification.** "Over time, it may make sense to codify these steps into laws or regulations. Regardless of whether this is required of companies… each of our companies are committed to doing this." **What's absent.** - No definition of superintelligence, frontier model, or capability thresholds. - No mention of model weights security, incident disclosure to government, compute reporting, or red-teaming standards (all present in the July 2023 Biden-era voluntary commitments and the 2024 Seoul Frontier AI Safety Commitments). - No named auditor standard (e.g., NIST AI RMF, ISO/IEC 42001) and no role for the US AI Safety Institute / CAISI. - No date, no effective period, no public reporting. **Signatories.** Six CEOs and one president. OpenAI signs through Greg Brockman (president), not Sam Altman (CEO). Musk signs for "XAI." Nvidia — a chip company, not a model developer — is included; Microsoft and Amazon, major model hosts and investors, aren't. **Terms.** - **Alignment** — whether a model's behavior matches developer intent; used here without definition. - **External evaluator** — third parties such as METR, Apollo Research, or government institutes that test model capabilities. - **Board committee** — the corporate-governance structure that public companies use for audit and risk; here applied to AI safety. ## 🗒️ Notes - The title promises "Super Intelligence"; the body is about operational controls for current frontier models. The mismatch is the most visible framing choice. - Layer (1) names the three standard catastrophic-risk domains — cyber, bio, chem — and adds "do not hack or access technical systems in unintended ways," which reads as an agentic-AI / autonomy concern. - Layers (2)–(4) are a classic three-lines-of-defense model from financial-services risk management: operations, internal oversight, independent assurance, board. - "Robust internal processes and controls" — "robust" is undefined. - "Partner with an independent external auditor or evaluator" — no criteria for independence (who pays, who chooses, what access). - "Together, these steps will give each company, its customers, and the public confidence" — the goal is stated as confidence, not safety outcomes. - "Each of our companies are committed" — subject-verb disagreement; the signature page's "Unites States" is a second typo. Small, but they suggest the document wasn't through standard White House editorial review. - The codification line is hedged twice ("Over time, it may make sense"). The accord doesn't commit the administration to anything. - Seven signatures on a page with the president's set apart on the left — a ceremonial layout. - Brockman, not Altman, signs for OpenAI. The doc gives no reason. ## 📚 References - White House post on X (source) — https://twitter.com/WhiteHouse/status/2105292669303791687 - White House, "Ensuring Safe, Secure, and Trustworthy AI" voluntary commitments (July 21, 2023) — archived at https://bidenwhitehouse.archives.gov/ - UK DSIT, "Frontier AI Safety Commitments, AI Seoul Summit 2024" — https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024 - NIST AI Risk Management Framework 1.0 — https://www.nist.gov/itl/ai-risk-management-framework - ISO/IEC 42001:2023 AI management systems — https://www.iso.org/standard/81230.html - Institute of Internal Auditors, "The IIA's Three Lines Model" (2020) — https://www.theiia.org/ - W3C, WCAG 2.2, Success Criterion 1.1.1 Non-text Content — https://www.w3.org/TR/WCAG22/#non-text-content ## 🗣️ Paraphrase - **Title.** The White House announces an accord on superintelligence. - **Preamble.** To build a good future, each company has to develop its technology safely and in ways that earn trust. - **Starting point.** Companies training frontier models need strong internal controls so models behave as intended and problems get fixed fast. On top of other precautions, each should adopt four layers. - **Layer 1.** Monitor model capabilities and alignment during training and deployment, especially for cyber, bio and chemical threats, and make sure models don't hack into systems. - **Layer 2.** Give an internal team the power to make sure the controls work and problems get fixed. - **Layer 3.** Bring in an independent outside auditor to check the controls. - **Layer 4.** Have an independent board committee oversee all of this and make sure issues get fixed. - **Result.** These steps give companies, customers and the public confidence the technology works as intended. - **Ongoing.** The companies will meet regularly to set standards. - **Future.** These steps may eventually become law; either way, the companies commit to them. - **Signatures.** The president, plus the heads of Google, Anthropic, Meta, OpenAI, xAI and Nvidia. ## 📰 Newsletter **Six AI companies signed a one-page safety accord. It's a JPEG.** On September 30 the White House posted the "White House Accord on Super Intelligence" to X: four layers of controls (internal monitoring, an internal team, an outside auditor, a board committee) signed by the president and the heads of Google, Anthropic, Meta, OpenAI, xAI and Nvidia. Two things WordPress folks will notice: - **The governance shape is familiar.** Internal team, independent review, oversight body — that's roughly how the WordPress plugin review and security teams work, minus the external auditor. The open question in both worlds is the same: who checks the checkers, and do they publish? - **The accessibility is not.** The entire accord was published as two images with `alt="Image"`. No text, no PDF, no web page found. That fails [WCAG 2.2 SC 1.1.1](https://www.w3.org/TR/WCAG22/#non-text-content) at the most basic level. If your plugin's AI features are going to cite government commitments, you'll be transcribing them yourself. Worth reading alongside the [2024 Seoul Frontier AI Safety Commitments](https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024) to see what's not in this one: no incident reporting, no weights security, no capability thresholds. ## ✅ Action Items #### DevRel - **Developer education:** when explaining AI-provider trust to developers, cite the actual text of commitments; this one exists only as images, so publish a transcription with a clear "transcribed, not official" label. #### OSPO - **Policy & Governance:** map the four layers to your own AI-use policy — do you have a named internal owner, an external review, and a board-level reporting line for AI features you ship? - **Security Management:** the accord's layer (1) names cyber/bio/chem plus "unintended access to technical systems"; that last item is the one relevant to agentic tools in developer workflows. Track how signatories report on it. - **Strategic Alignment:** treat the accord as non-binding; don't cite it as assurance in vendor reviews without company-specific evidence. #### Community - **Ecosystem:** open-source model developers and non-US labs aren't party to this; community AI policies shouldn't assume it covers the models in use. #### WordPress - **Core:** Not applicable directly. - **Core Performance:** Not applicable. - **Core AI:** the AI building blocks and Abilities API connect to models from several signatories. A short Make/Core AI note on what provider commitments do and don't guarantee would help plugin authors avoid over-claiming "safe AI." - **Core Program Team:** Not applicable. - **Mobile:** Not applicable. - **Playground:** Not applicable. - **Community:** Not applicable. - **Training:** Not applicable. - **Marketing:** Not applicable. - **Support:** Not applicable. - **Design:** Not applicable. - **Documentation:** Not applicable. - **Photos:** Not applicable. - **TV:** Not applicable. - **Polyglots:** Not applicable. - **Accessibility:** a teachable example of SC 1.1.1 failure from an official source. - **Meta:** Not applicable. - **CLI:** Not applicable. - **Hosting:** hosts offering AI features should not market them as "accord-compliant"; there's no compliance standard. - **Tide:** Not applicable. - **Openverse:** Not applicable. - **Themes:** Not applicable. - **Plugins:** plugin review guidelines on AI claims could note that voluntary provider commitments aren't certifications. #### The WPOCC - Not applicable to this source. #### FAIR - **Productized governance artifacts:** the four-layer structure (internal control, internal team, external auditor, board committee) is a usable template for a FAIR-style trust scorecard — and a reminder that without published criteria, "independent" is just a word. ## 🧪 Hypotheses 1. **H1.** Within 12 months, fewer than half of the six signatories will publicly name their external auditor or evaluator. 2. **H2.** No signatory will publish a board-committee report on AI safety controls within 12 months. 3. **H3.** The accord will not be codified into federal law or regulation within 24 months. 4. **H4.** Signatories' published safety frameworks (RSPs, preparedness frameworks) already meet all four layers, so the accord changes no company practice. 5. **H5.** Media coverage of the accord will cite the title's "superintelligence" framing more often than any of the four layers. ## 📊 Methodology - **H1/H2:** Track signatory websites, 10-K/annual reports (for the four public companies), and safety framework updates monthly; code for named auditors and board reports. - **H3:** Monitor Congress.gov and the Federal Register for bills or rules referencing the accord. - **H4:** Side-by-side comparison of Anthropic's Responsible Scaling Policy, OpenAI's Preparedness Framework, Google DeepMind's Frontier Safety Framework, Meta's Frontier AI Framework, and xAI's published policies against each layer. - **H5:** Content analysis of the first 100 news articles indexed after September 30, 2026. ## ❓ Questions - Is there an official text version anywhere (whitehouse.gov, Federal Register)? - Why does Brockman sign for OpenAI rather than Altman? - What counts as "independent" for the external auditor — who selects and pays them? - Why is Nvidia included and Microsoft and Amazon not? - What happened to the US AI Safety Institute / CAISI's role in evaluation? - Will companies report findings publicly, to government, or only to their own boards? - What does "super intelligence" in the title commit anyone to, given the body never mentions it? ## 🔮 Future Research - Comparative analysis of voluntary AI commitments (2023 White House, 2024 Seoul, 2026 accord) on specificity and follow-through. - Effectiveness of three-lines-of-defense governance in technology risk outside finance. - Accessibility of official government AI-policy communications. ## 💡 Implications - **Practice.** Companies already running RSP-style frameworks likely meet the letter of this accord; the test is whether they name auditors and publish reports. - **Policy.** The accord shifts responsibility to companies and floats, without committing to, future regulation. - **Security.** Layer (1)'s "do not hack or access technical systems in unintended ways" is the first explicit agentic-security commitment in a White House AI document; no measurement standard accompanies it. - **Education.** A clear case study in the gap between governance structure and governance substance. - **Funding.** External evaluation capacity (third-party auditors) will need funding and standards if layer (3) is to mean anything. ## 🧠 Learning Styles Assessment - **Reading level estimate:** Flesch-Kincaid Grade ≈ 14–15 (page 1 text; long sentences with nested clauses); Flesch Reading Ease ≈ 30 (difficult). #### 👀 Visual learners - **Present:** the scanned document and signature page. - **Improve:** a diagram of the four layers and who reports to whom. #### 🔊 Aural learners - **Present:** none. - **Improve:** an audio reading of the text (impossible from images without OCR). #### 🗣️ Verbal learners - **Present:** short, declarative commitments. - **Improve:** definitions for "alignment," "frontier model," "independent." #### 🧩 Kinesthetic learners - **Present:** none. - **Improve:** a checklist organizations could use to self-assess against the four layers. #### 🧠 Logical learners - **Present:** numbered layers. - **Improve:** criteria, metrics, timelines. #### 🧑‍🤝‍🧑 Social learners - **Present:** replies on X. - **Improve:** a public comment process. #### 🧘‍♀️ Solitary learners - **Present:** a one-page read. - **Improve:** Not applicable. #### 🌿 Naturalistic learners - **Present:** none. - **Improve:** Not applicable to this source. ## 📊 Accessibility Assessment #### Screen reader semantics - **Present:** the post's only text is the title. - **Improve:** both images have `alt="Image"`. A screen-reader user gets no access to the commitments or the signatories. This is the central accessibility failure. #### Keyboard navigation - **Present:** platform-dependent. - **Improve:** Not applicable to the author. #### Contrast and readability - **Present:** black serif text on white in the scan; adequate contrast. - **Improve:** low-resolution scan (611×791 px); text blurs when zoomed. #### Text resizing and reflow - **Present:** none — text in an image doesn't reflow. - **Improve:** publish as HTML. #### Media and captions - **Present:** no video. - **Improve:** Not applicable. #### Forms and interactions - **Present:** none. - **Improve:** Not applicable. #### Standards reference - **WCAG 2.2 AA 1.1.1 Non-text Content** — fails (`alt="Image"` on text content). - **1.4.5 Images of Text** — fails (text presented as image where HTML is possible). - **1.4.4 Resize Text / 1.4.10 Reflow** — fails as a consequence. - Also inconsistent with Section 508 obligations for federal ICT. ## 🌍 DEIB Assessment #### Cultural breadth - **Present:** US companies only; "the American people and the world" in the preamble. - **Improve:** no international lab or government signs; frontier AI is global. #### Inclusive language - **Present:** neutral. - **Improve:** Not applicable. #### Neurodiversity support - **Present:** numbered list. - **Improve:** Not applicable. #### Representation - **Present:** seven signatories: one head of state and six heads of large US tech companies. - **Improve:** no civil society, academic, labor, or public-interest signatory; the public is named as a beneficiary but has no role. #### Psychological safety - **Present:** Not applicable. - **Improve:** Not applicable. ## ✅ Logical Validity #### Strong reasoning - **Layered assurance structure.** Separating operational controls, internal oversight, external audit and board oversight is a well-tested model (IIA Three Lines). Strength: no single layer is trusted alone. - **Risk domains named.** Cyber, bio, chem and unintended system access are the domains most frontier frameworks prioritize. #### Fallacies - **Equivocation / title-body mismatch** — #fallacy/equivocation. Location: title "Accord on Super Intelligence" vs. body about current frontier model controls. Weakness: the title implies a commitment about superintelligence the text doesn't make. Fix: title it "Joint Commitment on Frontier Responsibilities," as page 1's subtitle already does. - **Vague terms carrying the argument** — Location: "robust," "independent," "operating as intended." Weakness: without criteria, compliance can't be tested. Fix: reference NIST AI RMF or ISO/IEC 42001 and define independence. - **Begging the question** — #fallacy/begging-the-question. Location: "these steps will give each company, its customers, and the public confidence that the technology is operating as intended." Weakness: assumes the controls work in order to conclude confidence is warranted. Fix: commit to publishing audit outcomes, which is what would justify confidence. #### Tone of voice - Formal, aspirational, corporate. Hedged where it touches law ("it may make sense"). Firm where it touches company responsibility ("each of our companies are committed"). ## 🛡️ DARVO Analysis - **Deny.** Not present. - **Attack.** Not present. - **Reverse victim/offender.** Not present. - **Overall.** Not applicable to this source — it's a joint commitment, not a response to accusation. ## 🎯 Conclusion - **Bottom line.** A one-page voluntary accord that sets out a sound governance skeleton (four layers, ending in board oversight) but leaves every key term undefined, attaches no reporting or enforcement, and is published only as inaccessible images. - **Next steps.** Look for an official text version; track whether any signatory names its external evaluator; compare against the 2024 Seoul commitments before citing this one as evidence of AI safety practice. ## 🔄 Refactor Appendix — White House Accord on Super Intelligence ### 📚 Literature Atoms #### [H1] Six AI CEOs sign a White House accord > [!quote] ID:rw_01m3sg3kv6cs3kgm8s7xwtzyxd > White House Accord on Super Intelligence **Context & use**: - The post's only text; all substance is in two images (transcribed in this analysis). - Anchor for the doc; the title's "Super Intelligence" doesn't appear in the body. **Tags**: #xref #share **Highlight**: inline (Phase 2) <!-- atomicity: 1 idea ✓ --> #### [H2] The accord requires an independent external auditor > [!quote] ID:rw_01m3sg3kv6cs3kgm8s7xwtzyxd > Partner with an independent external auditor or evaluator to carry out independent assessments of whether the controls, monitoring, and detection are operating as intended. **Context & use**: - Page 1 image, layer (3). Transcribed from the image; not present in Reader's HTML, so no inline anchor. - The most consequential commitment, and the least defined. **Tags**: #xref **Highlight**: not anchorable (image text) <!-- atomicity: 1 idea ✓ --> #### [H3] Codification is floated, not committed > [!quote] ID:rw_01m3sg3kv6cs3kgm8s7xwtzyxd > Over time, it may make sense to codify these steps into laws or regulations. **Context & use**: - Page 1 image, final paragraph. Transcribed; not anchorable. - Shows the accord is voluntary and the administration commits to nothing. **Tags**: #xref **Highlight**: not anchorable (image text) <!-- atomicity: 1 idea ✓ --> ### 🧠 Permanent Candidates #### [P1] governance-structure-without-published-criteria-is-unauditable **Claim** (one sentence, Courtney's voice): An oversight structure means nothing to outsiders until it publishes the criteria it checks against. **Evidence pointer**: [[#H2]] **Scope & conditions**: Applies to voluntary commitments, plugin review, and trust scorecards alike; doesn't apply where a regulator holds the criteria and enforces them. **Links**: - [[Using Claude Code Spending Your Effort]] — because contrasts on vendor self-evaluation - [[WordPress Takes Its Turn Leading the Open Website Alliance]] — because applies-in policy advocacy on AI and open web <!-- atomicity: 1 idea ✓ --> ### ✏️ Fleeting Queue - [F1] Find an official text version of the accord (whitehouse.gov, Federal Register) #fleeting #inbox - [F2] Compare against the 2024 Seoul Frontier AI Safety Commitments, item by item #fleeting #inbox ### → Things 3 - [ ] Check whitehouse.gov for an official text of the Sept 30 AI accord — X post was images only #things <!-- deep-read:sections end --> 📄 **Full text:** [[Full Document Contents/Tweets/White House Accord on Super Intelligence.md|White House Accord on Super Intelligence by The White House]] ## Highlights - White House Accord on Super Intelligence ([View Highlight](https://read.readwise.io/read/01m43fsmmb16q2x0jxwj6jj5d9)) - Note: [H1] Six AI CEOs sign a White House accord - Tags: atom/literature/h1