Trump Hosts AI CEOs; Six Frontier Firms Sign Voluntary "Constitution" — Then Renames AI "Super Intelligence" and Launches America.gov
Update (October 2, 2026): the first full dissection of the accord is out, and it confirms the enforcement gap. TechTimes' analysis of the "White House Accord on Super Intelligence: Joint Commitment on Frontier Responsibilities" (signed Sept 29 by Google, Anthropic, Meta, OpenAI, Nvidia and xAI) finds it contains "any penalty for companies that skip its commitments, any requirement to report AI safety incidents to any government body, or any requirement to use auditors the companies themselves did not select" — i.e., none of those things. The four commitments are internal: monitor capabilities/alignment during training and deployment (cyber, bio, chem), an internal team to confirm controls work, an external auditor of the company's choosing, and an independent board committee to receive reports. Nothing requires those reports to reach government, the public, or affected third parties.
The critique's sharpest point is that the accord's structure replicates exactly the arrangement that produced the three-month Medicare disclosure delay: "the company controls what its internal team surfaces, what its auditor reviews, and what its board committee receives... the same configuration that produced the delayed Australian disclosure is now embedded in the document described as 'morally binding.'"
What the dissection adds beyond the signing-week coverage:
- CAISI is the real teeth, and it's separate. The Commerce/NIST framework gives the government 30-day pre-release access to frontier models of five labs (OpenAI, Anthropic, Google DeepMind, Microsoft, xAI — expanded May 2026), with guardrails reduced or removed, 40+ evaluations completed. Meta signed the accord but has no CAISI agreement; Nvidia signed though it builds no frontier models. The original June 2026 EO proposed 90-day review; David Sacks helped cut it to 30 and bar any licensing regime.
- The oversight logjam is documented: OpenAI has told Congress it is building automated shutdown capabilities but has withheld Hugging Face breach logs demanded by the August 10 House oversight letter led by Rep. Greg Casar (31 Democrats) — who calls the refusal "deeply concerning."
- The Kill Switch Act wouldn't have caught the Medicare breach either — the Lieu-Moran bill explicitly excludes red-teaming and structured testing from its covered-incident definition.
- Context from signing week: Trump simultaneously signed an EO rebranding "AI" as "Super Intelligence/SI" in official documents (which briefly broke America.gov and drove a 20x spike in Slovenia's .si domain registrations), and Google's restricted Argon release the next day showed capability-gating is now the labs' own preferred answer to the same problem.
Why it matters: the accord's defenders call it a constitution; its structure is a press release with a board committee. The test the TechTimes piece sets up — does any signatory report the next incident faster than OpenAI reported Medicare? — is now the operating question for SAFA: The Labs' FINRA-Style Self-Regulator Takes Shape — Without Government, Meta, xAI or Nvidia, the FTC probe (The Rogue-AI Enforcement Wave Goes Multi-Front: FTC Probe Now Joined by California AG Subpoena and a 15-State Coalition), and the Oct 5/6 sworn testimony.