The Fifty-Percent Hedge: A Senator Tries to Outlaw a Superlative
On September 11, 2026, Senator Bernie Sanders went on BBC Newsnight to promote legislation banning artificial superintelligence and pausing advanced AI development — and, in the same breath, a sovereign wealth fund in which the United States would take a 50% stake in AI companies.12 “When scientists tell you there is a chance, a chance that it could have a cataclysmic impact on humanity, you’ve got be a moron not to say, slow it down.” The proposal arrived on a news cycle built for it: that same day Anthropic published a threat-intelligence report detailing blocked attempts to use Claude for biological-weapons development;3 the same week saw Evan Hubinger concede a greater-than-10% chance AI kills everyone this decade, OpenAI’s chief scientist call for voluntary slowdowns, and an open letter demanding a superintelligence treaty.4 President Trump rejected the premise within hours: “if we don’t win AI, we’re going to be put in a very bad position.”1 Five frames on the first legislation to take the labs’ own risk estimates at face value — and to embed a paradox in its financing.
1. First Principles — what would a ban consist of?
A workable prohibition needs three components: a definition of the banned object, a detection mechanism, and an enforcement point. Superintelligence offers none. There is no agreed definition — the thing does not exist yet, so the bill outlaws a superlative, not a technology. There is no detection test — capability is latent, and a frontier checkpoint looks byte-for-byte like any other tensor file. And there is no natural chokepoint — weights are files, compute is distributed, training runs are private. Rebuilt from fundamentals, a state can only regulate activities: compute thresholds, training-run registrations, release events. The difference from the default answer is stark — as described, the statute bans an adjective and waits for a noun to violate it.
2. Golden Circle — the why says never, the how says half
Why (stated): prevent a cataclysm its own builders price at >10%. How: pause advanced development, ban the endpoint, and take 50% equity. What: legislation plus a sovereign wealth fund. The alignment check fails at the How. A why that says never build the thing cannot be served by a how that owns half of whatever gets built. A ban plus a stake is not a policy; it is a hedge — long the catastrophe it forbids. The inside-out version tells the truth: the state cannot decide whether to prevent this industry or nationalize it, so it is doing both.
3. Nietzsche Ladder — who wrote the tablets
Camel. The burden is real and honestly carried: decades of precautionary warnings, the scientists’ letters, a chief scientist begging for voluntary restraint, a researcher staking his name on a decimal. Sanders is not inventing the fear; he is the first legislator to treat it as admissible.4
Lion (responding to Camel). Yes, those are the tablets; now ask who wrote the financing section. A 50% stake converts prevention into position — a state that owns half of AI profits when AI accelerates, and every dollar of that upside is a reason not to enforce the ban. The tablets say slow down; the ledger says speed up. And across the aisle sits the mirror image: Trump’s “win AI” is the same race with opposite signage.
Child (responding to Lion). Your No is necessary, but negation is not governance. What could be built instead: “slow it down” only means something if someone can measure the speed — compute thresholds with verification, training-run disclosure, release gates with published criteria, and a public stake held in escrow against safety compliance rather than as open-ended ownership. Fear becomes law only when fear is given instruments.
4. Analogy Transfer — the verification gap
Structural form: how does a system halt development of a technology whose danger is probabilistic, whose artifacts are intangible, and whose development is driven by competition? Nuclear arms control solved it with countable warheads and satellite verification. FDA clinical holds pause drug programs with defined release criteria. Market circuit breakers halt trading with pre-agreed restart conditions. The strongest transfer is the circuit breaker’s pre-agreed restart — a pause that specifies what evidence resumes it, which a bare ban never does. But the disanalogy check is brutal: warheads can be counted from orbit and trials have endpoints; model capability is latent and unverifiable. The analogy breaks exactly where the ban needs it to hold. A superintelligence treaty without a detector is a armistice without observers.
5. Inversion — how to guarantee the ban fails
Invert the goal. To guarantee failure: (1) define “superintelligence” vaguely, so enforcement becomes discretionary — High likelihood, High damage; (2) ban it US-only, so the race moves offshore to actors with worse safeguards while domestic labs bleed talent — High, High; (3) fund the government with 50% of the upside, so every enforcement decision carries a moral-hazard tax — Medium, High; (4) write it for the midterms rather than the statute books — High, Medium, and entirely self-inflicted. The guards negate into a program: measurable capability thresholds, treaty-first sequencing with allies (the Burnham letter shows the appetite), escrowed rather than open equity.
Synthesis
First principles finds a law regulating a word. The Golden Circle finds a hedge between prevention and nationalization. The ladder finds state capture hiding inside the safety tablets. Analogy transfer finds the ban dying at the verification gap, and inversion enumerates its failure modes — offshore flight, discretionary policing, profiting from the peril. The frames converge on one finding: for years the labs priced the risk and asked to be trusted with it; Sanders is the first actor to accept their number and demand the corresponding policy. What he has not done — what no one has done — is build the instruments that made every prior two-handed policy work: the metric, the detector, the treaty. The state can now convert fear into legislation. Converting fear into enforcement is still unsolved, and the 50% stake quietly bids for the other side.