Thursday links: Prediction markets, agent hackers, quantum risks

This is a segment from The Breakdown newsletter. To read full editions, subscribe.

“Even when I was little, I was big.”

— William Perry

Michael Lewis dates the beginning of betting on everything to 1985 when Caesars Palace offered 20-1 odds on William “Refrigerator” Perry scoring a touchdown in the Super Bowl.

He did — a one-yard rumble into the end zone — and Caesars lost at least $250,000 on the bet (a substantial sum in an era when bettors had to call Las Vegas by phone to place wagers).

That was the first known “proposition” bet — a bet on anything other than the score or result of a game, and as such, the primordial ancestor to today’s prediction markets that offer odds on seemingly everything.

Despite the loss it took, the president of Caesars called it the best bet they’d ever taken: “We’ve had so much publicity on the bet, whatever we’ve lost was worth it.”

“The next day, just about every book in town, your phone started to light up from every big city in the country,” the bookmaker Jimmy Vaccaro told Lewis. “They heard about Perry scoring a touchdown.”

Now, bookmakers “write” more business on the props — will the coin toss be heads or tails? what will the first commercial be for? will there be a wardrobe malfunction? — than they do on the games.

Before The Fridge, there were only about three bets to make on the Super Bowl: the winner, the total points, the halftime score.

Now, there are hundreds.

As it turned out, that was the last of The Fridge’s three career touchdowns despite his 10 additional years in the league (playing on defense).

But those three short rumbles set us on the path to today’s world of betting on virtually anything with prediction markets.

(Although we still mostly want to bet on sports.)

A look inside political markets

Here’s a measure of how sophisticated prediction markets have gotten: When Andy Hall watches sports, he’s now gotten in the habit of checking prediction markets right before a big play “because the markets move several seconds before my TV feed does.”

Hall teaches political science at Stanford, so he’s more interested in how prediction markets are “changing how we monitor and understand politics.”

Newsletter

Subscribe to The Breakdown

Hall seems cautiously optimistic that booming prediction markets will prove to be a public good because they should offer “a clearer shared picture of a highly complex political environment.”

But he’s also concerned about the “strange feedback loops” they create.

As an example, he cites a Virginia Attorney General race where unverified claims about exit polls circulating on social media moved prediction markets, which caused social media to post the “BREAKING NEWS” of the move in prediction markets, which, in turn, caused the prediction markets to move even further.

That probably did not do the public any good.

Hall notes that prediction markets are also raising questions about what it means to “win” an election.

“In a world where prediction markets are increasingly seen as sources of truth — many people pointed to when Kalshi ‘called’ the mayoral election in New York City as the proof that Mamdani had won — determining these edge cases will be exceedingly fraught.”

So what does it mean if Kalshi calls a close presidential election in 2028?

I’m not sure I want to find out.

(For another foreboding example of prediction markets affecting reality, see this story about bettors changing maps of the frontline in Ukraine.)

AI agents are very good at hacking smart-contracts

Researchers at Anthropic report that their AI agents successfully exploited 56% of smart contracts known to have vulnerabilities.

More impressively (or worryingly), when tested on 2,849 smart contracts with no known vulnerabilities, the agents also discovered two novel “zero-day” exploits — a demonstration that agents can find new vulnerabilities autonomously.

Perhaps most worryingly, the agents are getting better at an astonishing rate: “Over the last year, frontier models’ exploit revenue…doubled roughly every 1.3 months,” the researchers report.

(They want you to know that no blockchains were harmed in this experiment — they tested on “blockchain simulators.”)

To put that in perspective, Moore’s Law describes the performance of semiconductors doubling every two years.

These AI agents are doubling their capacity to exploit smart contracts every six weeks.

Amazing. And terrifying.

Anthropic does not have a particular interest in crypto. Instead, it has run this experiment to measure how adept agents will be at hacking code of any sort: “The same capabilities that make agents effective at exploiting smart contracts — such as long-horizon reasoning, boundary analysis, and iterative tool use — extend to all kinds of software.”

Here’s another amazing thing it’s found: “It costs just $1.22 on average for an agent to exhaustively scan a contract for vulnerability.”

“As costs continue to fall,” it concludes, “attackers will deploy more AI agents to probe any code that is along the path to valuable assets, no matter how obscure.”

Consider investing in a lockbox to hold all the physical cash and gold coins we’ll soon be back to using.

Quantum computers are coming for crypto first

The risk that quantum computers pose to crypto is often dismissed with a hand-wavy assertion that once the computers get that good, everything will be at risk, so we’ll have bigger things to worry about.

But the latest episode of Epicenter explains why crypto is uniquely vulnerable.

“Most of the classical cryptography community…has been working on this for years,” Stefano Gogioso explains. “They already have protocols they could use if quantum were to happen tomorrow.”

In Web2, “you have centralized authorities,” he adds. If worse comes to worst, “you have banks that would refuse to transact for a day or two.”

In crypto, however, there are two problems: 1) blockchains use “new applications of cryptography” that have gotten little attention from researchers studying quantum resistance and 2) even if we had solutions, there are so many different voices, opinions and vested interests in decentralized crypto, “it would be challenging to change the infrastructure.”

Even more fundamentally, Gogioso notes, “we built an entire ecosystem for the very purpose of making [our data] available to everybody at all times” — which means that once quantum computers are available, they’ll be able to decrypt the entire transaction history of every blockchain.

So, if you’ve been using a privacy chain like Monero to make illicit transactions, you might want to relocate to a non-extradition country sometime soon.

“Monero is already broken,” Gogioso says.

Gogioso also expects governments to use quantum computers to identify who hasn’t been paying capital gains taxes on their crypto winnings: “They’re certainly going to send them a bill for the money they haven’t been paid yet.”

(Personally, I look forward to the refund I’ll get on all the crypto losses I’ve been too lazy to report.)

Other fun scenarios include someone moving one of Satoshi’s coins — how will we know if that’s Satoshi himself or just a guy with a quantum computer?

Or consider a US rival like China targeting Bitcoin — not to make money, but to destabilize a system that’s become ingrained in US finance.

But the biggest threat of all might be Gogioso’s final prediction: that quantum technology could create a form of permissionless, digital money that solves the double-spend problem without resorting to crypto’s exploit-prone model of distributed consensus.

Whatever the risks, the response has been woefully inadequate.

John Lilic notes that because crypto has a $3 trillion market capitalization, if there’s even a 1% chance of total loss, “we should be spending $30 billion” on making crypto quantum resistant.

His assessment of the current investment? “We’re not spending anything.”

Get the news in your inbox. Explore Blockworks newsletters:

The Breakdown: Decoding crypto and the markets. Daily.
0xResearch: Alpha in your inbox. Think like an analyst.