Skip to main content

On Thursday 19 November 2026 the United States Senate passed S. 4035, the Discouraging Exploitative Assassination, Tragedy, and Harm Betting in Event Trading Systems Act, which everyone sensibly calls the DEATH BETS Act. It went through by voice vote after a substitute amendment was adopted by unanimous consent. It now goes to the House. If it becomes law, no CFTC-registered exchange may list or clear a contract on terrorism, assassination, war, US military operations, or the death of a person.

The headlines say Congress banned death bets. Read the engrossed text against the version Senator Adam Schiff introduced in March and something more interesting turns up. The bill that passed does not test what the bill that was introduced tested. In March the trigger was any contract that “involves, relates to, or references” those events, or that “could otherwise be construed as correlating closely to an individual’s death”. The text that passed on Thursday catches contracts that “expressly reference” a listed event, plus contracts whose event, “at the time of listing, is reasonably expected to resolve principally based on a death of an individual”.

Correlation became vocabulary. A question about the world became a question about the wording and an expectation frozen at one moment. I do not think that was a cynical move. Correlation is vague and the drafters needed something an exchange’s compliance team could actually apply. But it means the law now identifies a dangerous contract by what it says, not by what it pays for. And the thing that makes a death bet dangerous was never what it says. It is what it pays for.

TL;DR

  • The Senate passed the DEATH BETS Act (S. 4035) on 19 November 2026 by voice vote; it bans CFTC-registered prediction markets from listing contracts on war, terrorism, assassination, US military operations and individual deaths, and now heads to the House.
  • The amendment narrowed the test from contracts “correlating closely” with a death to contracts that “expressly reference” a listed event or are expected “at the time of listing” to resolve principally on a death.
  • Every open wager is a standing bounty for whoever can cause its outcome, sized by the money on the other side. That bounty is the real harm, and it exists for any outcome a small group of people can move, whatever the contract’s wording.
  • Chainlink’s own VRF security guidance is the one place in the on-chain gaming stack that tells builders to size against “value-at-risk”. In an eight-day sample of 15,740 Base VRF draws, 83.65% asked for the minimum of zero confirmations.
  • On-chain games can do what the Senate cannot: read the pot as a balance, list every role that can touch the outcome, and check the price of moving it, block by block. That is the standard Satoshie is built to.

This is the hundred-and-sixth unasked half of fairness.

What the sponsor said the harm was

When Schiff introduced the bill in March he gave three reasons in one sentence: “Betting on war and death creates an environment in which insiders can profit off of classified information, our national security is jeopardized, and violence is encouraged.”

The coverage has been almost entirely about the first. That is understandable. The bill was introduced after on-chain analysts flagged a cluster of freshly created Polymarket wallets that reportedly made around $1 million betting on the timing of US strikes against Iran, several of the positions placed hours before the first reported explosions. That is an insider-information story, and this blog has already written its fairness half: in The Outcome Somebody Already Knew the argument was that insider trading is only possible against an outcome that already exists somewhere before settlement, and is not merely harder but impossible against a random number that has not been generated yet.

The House is getting the same theme from another direction. Representative Sean Casten has proposed a whistleblower programme paying 10% to 30% of sanctions in prediction market fraud cases above $1 million. Hold on to that one, because it comes back later.

The third harm in Schiff’s sentence is a different animal. “Violence is encouraged” is not about anyone knowing the outcome. It is about someone making it happen. An insider trades on information. The person this clause worries about trades on capability. Those are separate failures, they need separate defences, and the second one is far older than prediction markets.

Every wager is a bounty

Strip a bet to its mechanics and it is a conditional transfer: if the event happens, money moves from the people who bet against it to the people who bet on it. Now look at that from the point of view of someone who can cause the event. To them an open market is not a forecast. It is a standing offer of payment, posted publicly, funded by strangers, and the size of the offer is exactly the money sitting on the other side.

Crypto has had a name for the extreme version since the mid-1990s, when the cypherpunk writer Jim Bell published an essay describing an “assassination market”: an anonymous pool that pays whoever correctly “predicts” the date of a named person’s death. The predicting is a fig leaf. The only bettor who can reliably win is the one who arranges the outcome, so the pool is a contract killing with a crowdfunded fee. The idea has been a fixture of crypto’s darker corners ever since, which is part of why a Senate bill about it passed without a recorded vote.

But nothing in that mechanism is specific to death. The bounty exists for any outcome where three things are true:

  • The pot is large. The bounty is the losing side’s stake, so it grows with open interest.
  • The movers are few. A small, identifiable group of people can change the result: a referee, an official who controls a date, a company officer who decides whether to announce.
  • Moving it is cheaper than the pot. If the cost of causing the outcome (effort, risk, legal exposure, reputation) is lower than the payment on offer, the offer is live.

Death and war sit at one end of that range because the cost of moving the outcome is a serious crime, and society would rather not find out how large a pot has to be before somebody pays it. A contract on whether a referee shows a red card in the first ten minutes sits at the other, and sport has a long and grubby history showing that a cheap outcome with a single mover gets fixed at surprisingly modest prices. Same structure. Different price tag.

What the passed test measures

Put the passed text next to that list. Clause (1) asks whether the contract “expressly references” terrorism, assassination, war or US military operations. That is a question about the words. Clause (2) asks whether, “at the time of listing”, the event is “reasonably expected to resolve principally based on a death”. That is a question about an expectation, assessed once, when the contract goes live.

Neither clause asks how large the pot is. Neither asks who can move the outcome or what it would cost them. Two contracts with identical wording can carry bounties a thousand times apart, and the law treats them identically. Two contracts carrying identical bounties can be worded differently, and the law treats them differently.

Clause (1)(E) does let the CFTC add “any similar activity” by rule, which is a sensible safety valve, and I would rather have a clear, enforceable ban on the worst category than a vague one nobody can apply. A Kalshi spokesperson told Fortune the company has never listed death markets at all, and Polymarket pulled a nuclear-detonation market earlier this year after a backlash. Nobody serious is defending the contracts this bill bans.

The point is narrower. The amendment also added a rule of construction stating that nothing in the Act changes the legal status of other event contracts, including the wider category of gaming contracts. So the law draws a line around a vocabulary list and explicitly declines to say anything about the property that makes items on the list dangerous. Every outcome with a large pot and a cheap, small set of movers that does not use the listed words is exactly where it was on Wednesday.

Which is why the House is also looking at a whistleblower bounty. If you cannot size the bounty a market creates, the fallback is to post a bigger one for whoever reports the person collecting it. That is a reasonable thing for a legislature to do. It is also an admission that the market itself cannot tell you who its movers are.

The same bounty sits on every game of chance

Bring this back to on-chain gaming, because games of chance have the same shape with one important difference.

A raffle, a coinflip or a spin is also a conditional transfer, and the pot is also a standing bounty for whoever can move the result. The difference is where the result comes from. A prediction market settles on an event out in the world, so its movers are out in the world too: soldiers, officials, referees, executives, anyone with a gun. You cannot list them, which is why legislators end up listing topics instead.

A game of chance generates its own outcome. Nobody outside the system can cause a coinflip to land heads, because there is no event in the world for them to influence. The movers are all inside the machinery: whoever produces the random number, whoever can ask for another one, whoever can supply the number in its place, whoever controls when a bet is accepted relative to when the number exists, and whoever can rewrite the block it landed in.

In a traditional casino that machinery is a server you will never see, so the movers are “the house” and the defence is a licence and an occasional lab certificate. Notice that this is the Senate’s approach again. A licence is a vocabulary test for operators: it tells you what a company is called and where it is registered, not who can touch which outcome today.

On-chain, both halves of the bounty are readable. The pot is a balance. You can query it. The movers are roles and functions in bytecode. You can list them. The price of moving the outcome is set by what those functions allow and what the chain costs to rewrite. You can check it, for this game, at this block, without asking anyone’s permission.

What VRF removes, and what it leaves for you to size

Chainlink VRF removes the biggest mover from the list. The oracle that produces the number cannot choose it: the output comes with a cryptographic proof, the coordinator contract checks that proof on-chain before the number is usable, and a forged or grinded value does not verify. That turns “the house picks” from a capability into an impossibility, which no licence has ever done.

It does not remove every mover, and Chainlink is unusually candid about the rest. Its VRF v2.5 security page is, as far as I can find, the only document anywhere in the on-chain gaming stack that tells builders to size their defences against the money at stake. Four items on it map directly onto the bounty list:

  • Chain rewrites. Validators “could rewrite the chain’s history to put a randomness request from your contract into a different block, which would result in a different VRF output.” They cannot pick the number, only “re-roll the dice”. The defence is the request’s confirmation count, chosen “to make potential rewrite attacks unprofitable in the context of your application and its value-at-risk”. Value-at-risk is the bounty, named explicitly.
  • Re-requests and cancellations. “Any re-request or cancellation of randomness is an incorrect use of VRF v2.5”, and builders “must prevent the ability for any party to discard unfavorable randomness.” Whoever can discard a result is a mover with an unlimited number of tries.
  • Late inputs. “Don’t accept bids/bets/inputs after you have made a randomness request.” Anyone who can still change their position after the request goes out is betting against a number that is closer to existing than everyone else’s.
  • Reverting fulfilment. If the consumer’s fulfillRandomWords reverts, the service does not try again. A contract that can be made to revert on an unwelcome result has quietly handed someone a veto.

Every one of those is a design choice in the game contract, not in VRF. VRF makes choosing the number impossible. Whether anyone can throw it away, ask again, bet late or veto it is up to whoever wrote the game.

Nobody sized it

This blog has measured how often that sizing actually happens, and the answer is: almost never.

Confirmations are the one parameter Chainlink explicitly ties to value-at-risk, and the game chooses it per request. In 13,166 Draws Asked for Zero Confirmations I read every request on the Base VRF 2.5 coordinator across eight days in October: 15,740 draws, and 83.65% of them asked for zero, the minimum allowed. Only five of the 201 legal values were used at all. In practice the delivered depth was never below what was requested (the zero-confirmation cohort was answered at a median of two blocks), so nobody was robbed by a rewrite that week. But a parameter the documentation says to set from the size of your pot was, for most draws, left at the floor regardless of the pot. That is not a sized bounty. It is a default.

The bigger finding was the other kind of mover. In The Prize Cap Was Committed First, 45 of the 49 game contracts sharing one Base VRF subscription carried a function, s_ChooseWinner(uint256,uint256[]), restricted to the operator, that runs the fulfilment logic on random words supplied by the caller. Whatever the operator’s intentions, that is a role that can settle a draw without the oracle’s proof, sitting next to a pool of player money. In bounty terms it is a single mover whose cost of moving is one transaction. Every one of those games advertises VRF, and the advert is true. It is also a vocabulary test.

None of this needed a subpoena, a whistleblower or an enforcement action to find. It needed an RPC endpoint and some patience. That is the actual advantage on-chain gaming has over every other kind of wager, and the industry mostly does not use it.

The Satoshie position

Satoshie’s raffles and coinflips use Chainlink VRF for the same reason everyone else’s do: the house should not be able to pick the number, and a proof beats a promise. But “we use VRF” is the expressly-references clause of on-chain gaming. It tells you which word appears in the documentation. The question that matters is the one the Senate’s text does not ask: for this pot, today, who can move the outcome, and what would it cost them?

The standard Satoshie is built to is that the answer should be “nobody, at any price”, and that you should be able to check that yourself instead of taking it from a blog post. That means no path that discards or re-requests a draw, no path that settles on a caller-supplied number, no entries accepted after the request goes out, a fulfilment that does not revert on an unwelcome result, and a pot that sits on-chain where anyone can read its size. A game built that way does not need a list of banned topics, because there is no event in the world for anyone to influence and no role in the contract that can do the influencing.

The limits of the claim

Three honest caveats.

First, none of this helps a prediction market. The whole value of a contract on a real-world event is that it settles on the real world, and the real world’s movers cannot be enumerated in bytecode. The DEATH BETS Act is aimed at a problem on-chain games simply do not have, which is not the same as on-chain games having solved it.

Second, listing the movers is not the same as removing them. Base still has a single sequencer, so the deepest rewrite risk on that chain is a question about one operator, and confirmation counts only price ordinary reorganisations. Upgradeable contracts can add a mover tomorrow that was not there today. A clean list is a reading at a block height, and it needs reading again.

Third, a game with no mover can still be a bad game. Fair odds and fair settlement say nothing about whether the expected return is decent or whether the operator will be around next year. A sized bounty answers “can anyone cheat this draw?” and nothing else.

Three questions for any game of chance

  1. How big is the bounty right now? Find the pot on-chain. Not the marketing figure: the balance the payout comes from, at the current block.
  2. Who can move the outcome? List every role and function that can request, cancel, re-request, supply, settle or veto a draw, and every key that can upgrade the contract to add one.
  3. What does moving it cost them, compared with the pot? If the answer is “one transaction” for any of them, the game is offering a bounty, whatever its documentation says.

The Senate looked at the most horrifying markets humans have invented and wrote down their names. That is probably the best a legislature can do with outcomes that happen out in the world. Games of chance do not have that excuse. Their outcomes are made inside a machine you can read, which means the bounty can be sized and the movers can be counted. Somebody just has to bother.

📷 Photo by Afif Ramdhasuma on Unsplash

Valentina Ní Críonna

Author Valentina Ní Críonna

More posts by Valentina Ní Críonna