Trump Orders Federal Agencies to Say Super Intelligence as Tech CEOs Sign Safety Pledge
President Donald Trump signed an executive order on September 29, 2026, directing federal agencies to call AI "super intelligence," alongside a voluntary safety accord signed with tech executives including Musk, Zuckerberg, Huang and Amodei.
The White House convened major tech CEOs in one room to sign an AI safety pledge. President Trump announced the commitment in a post on Truth Social. He called the pledge signed by tech leaders "morally binding." TechCrunch
Those present included Elon Musk and Anthropic CEO Dario Amodei. Mark Zuckerberg was also among the CEOs convened, representing Meta as a signatory to the commitment. Fortune The White House accord itself was signed alongside Musk, Zuckerberg, Huang and Amodei. Forbes
The executive order is titled "Inaugurating The Era Of Super Intelligence" and was published on September 29, 2026. It directs executive departments and agencies to use "Super Intelligence" and "SI" to the maximum extent permitted by law. The text is short. The mechanism is definitional, so it changes the official label rather than the technology itself.
The White House published a companion document titled "Fact Sheet: President Donald J. Trump Inaugurates the Era of Super Intelligence" about the order. White House That fact sheet sits alongside the March 2026 unveiling of a National Super Intelligence Policy Framework, which provided the earlier policy framing for the terminology shift.
Separate from the nomenclature order is the White House Accord on Super Intelligence. It is 308 words long. It includes four layers of controls and audits adopted voluntarily by companies. Forbes The accord is a voluntary pact urging companies to self-police on AI safety. Al Jazeera
Both Dario Amodei and Elon Musk have warned that unchecked AI could pose an existential risk to humanity. Their presence as signatories puts two of the most explicit voices on catastrophic risk inside a voluntary framework rather than outside it.
The broader context here is worth spelling out for engineering and compliance teams. A federal naming rule does not change model architectures, training runs, or inference stacks, the systems used to design, train, and run models. It does change paperwork. Federal filings, grant language, procurement documents, audit reports, and incident disclosures will now be expected to say "Super Intelligence" and "SI" where they previously said AI. For platform teams that maintain model cards, evaluation harnesses, logging pipelines, and access controls, the records that describe models, test them, track activity, and limit access, terminology mapping becomes a real maintenance task.
In my view, the more consequential technical question is what the four layers of controls and audits will mean in practice. Voluntary adoption leaves the details to each signatory. Experienced practitioners will recognize the gap between self-attestation, where a company grades its own work, and independent verification. Controls can mean pre-deployment evaluations, red teaming, which is structured attack testing, deployment gating, and post-deployment monitoring. Audits can mean internal review, second-party assessment, or third-party inspection with access to weights, data lineage, and eval results, the core files, data history, and test scores behind a model. A 308-word accord cannot resolve those distinctions.
Worth flagging, the phrase "morally binding" is doing specific work. It signals CEO-level commitment without creating a new enforcement instrument. For risk officers, that distinction matters. Moral commitment can focus internal resources and unblock safety roadmaps. It does not replace contract terms, export controls, liability regimes, or formal rulemaking. Teams should treat the accord as a statement of intent to be tracked in governance registers, not as a control framework that satisfies an external audit on its own.
Looking at what this enables, there is reason for measured optimism. Getting competing lab and platform chiefs into one room to sign the same safety text, however brief, establishes a common reference point. Shared vocabulary and a shared acknowledgment of responsibility make interoperable practices easier to build. If the four audit layers evolve into testable checklists with common eval suites and disclosure formats, enterprise adopters, cloud providers, and security reviewers will benefit from less fragmentation.
