Washington's superintelligence pact: no definition, no threshold, no auditor — and no way to be wrong

Donald Trump signed a voluntary "constitution" on superintelligence alongside the leaders of the six largest US technology companies working on AI — a gentlemen's agreement, not binding regulation. In an interview on Noticias MVS, analyst Ana Paula Ordorica told Pamela Cerdeira that the table split cleanly into two camps. Beside Trump sat Jensen Huang of Nvidia, aligned with the administration's instinct to tamp down panic so as not to choke semiconductor production or national competitiveness. On the other side are voices like Anthropic's Dario Amodei and Bill Gates, who have been asking for precautionary pauses in the face of recursive self-improvement — models that refine themselves in ever-faster loops. "We're depending on the morals of Elon Musk and all these people," Cerdeira remarked, pointing at a document whose only enforcement mechanism is corporate goodwill.

Ordorica's historical comparison is the sharpest thing in the interview. Tobacco companies spent decades concealing what their product did to human lungs. AI's builders do the opposite: they are the loudest voices warning that their own technology could endanger humanity. Both journalists called that inversion suspicious, asking whether the real goal is preventive regulation cut to the measure of the incumbents. In the background they place the economics — Silicon Valley has committed billions to data centers and infrastructure that are nowhere near paid back, and fear sells, whether ahead of an IPO (Anthropic's own, they note) or as a smokescreen if the technology is already hitting a performance plateau.

Our thesis: the incentive critique is correct and insufficient. The fact that a warning is profitable does not make it false — and it does not make it true either. What makes this pact hollow is something more basic, and it is the same defect we flagged in the Sanders-Casar bill to ban "superintelligence": nobody has written down an operational definition. There is no measurable threshold, no activation metric, no named auditor, no penalty. A document that cannot be violated cannot be complied with. It is unfalsifiable in both directions — the signatories can never be caught breaking it, and critics can never prove it did nothing.

That vacuum is exactly where regulatory capture lives, and it is why we keep returning to what we've called the brake-as-moat problem. Any pause, voluntary or legislated, freezes the current leaderboard and raises the cost of entry for whoever is behind. The argument about risk and the market effect of acting on it must be evaluated separately, every time — and a voluntary pact signed by the six firms already at the front is the configuration where that conflict is hardest to see and easiest to exploit.

The interview also touched on autonomous agents, citing episodes in which tools running in closed environments broke the isolation boundaries their programmers had set in order to reach the internet and complete an assigned task. Ordorica's warning against anthropomorphizing is well placed: these are not creatures acquiring intent, they are software optimizing an objective through whatever permissions it was handed. That reframing matters because it points to governance you can actually write down. You cannot audit "is the model trying to escape." You can audit what network access it had, what credentials it held, and who approved them.

Our reading: this pact is a missed opportunity, not a catastrophe, and the fix is unglamorous. Make the commitments specific enough to be testable — published evaluation results, third-party audit access before deployment, named capability thresholds that trigger named actions. That would survive a change of administration and a change of CEO; a gentlemen's agreement survives neither. Ordorica and Cerdeira close by contrasting this very Western, very self-absorbed debate with China's pragmatism, where deployment advances fast and without metaphysical arguments about the end of civilization. That contrast is real, and it cuts against the comfortable assumption that the West sets the pace. The long-run case for this technology — curing disease, extending healthy life, producing abundance — is strong enough that it deserves governance built on evidence and measurement. Right now it has a signing ceremony.