- The Breakdown
- Posts
- 🟪 Deep fakes, home invaders, and sanctions
🟪 Deep fakes, home invaders, and sanctions
Will AI make financial markets more or less efficient?


Thursday links: AI, deep fakes, home invaders, insiders, and sanctions
Will AI make financial markets more or less efficient? A recent study answers “yes.”
Researchers used GPTZero to detect AI-generated content on two platforms, Seeking Alpha and WallStreetBets, and then measured how that content affected the stock market.
Incredibly (to me, at least), the AI content on Seeking Alpha made markets more efficient: “following AI deployment, bid-ask spreads narrow and retail order flows become more informative.”
AI research makes markets better and less noisy. Who’d have thunk it?
Not so incredibly, the AI content on WallStreetBets made markets less efficient: “AI deployment is followed by higher abnormal trading volume, greater volatility, and a higher incidence of lottery-like return outcomes, consistent with AI amplifying speculative dynamics and sentiment-driven trading.”
AI shitposting makes markets, well, shittier. I would have thunk that, for sure.
The authors think this teaches us something about AI more generally, as well: “The nature of AI-assisted thought transmission… is not inherent in the technology itself,” they conclude.
In other words, AI will make the world both smarter and dumber, depending on the environment it finds itself in.
Should be fun.
“Anger swelled in a poor Bolivian city where a plane crash left banknotes strewn on the ground,” Reuters reports, “prompting authorities to burn the notes they collected and use tear gas to disperse those they accused of trying to pick them up.
The plane was carrying 17.1 million banknotes of various denominations, totaling 423 million bolivianos (worth roughly $59 million).
"Look at all the money they burned,” a local told Reuters. “There are poor families that need that money."
Had the government also dispatched a monetary economist to the scene, they could have explained that the destroyed banknotes did not represent wealth, that releasing them into circulation would expand the money supply without increasing real output, and that upward pressure on prices would erode everyone’s purchasing power.
I'm confident that would have peacefully dispersed the crowd.
The internet is flooded with fake war videos, including a highly convincing one depicting missiles hitting Tel Aviv that racked up 20 million views on X. Grok has been adding to this fog of war by telling users some fake videos are real when they respond to posts with “@Grok is this real?”.
Cryptography in every phone might fix this.
Brevis Vera is a media-provenance protocol that uses zero-knowledge proofs to verify that media originated from a trusted device. Enabled devices cryptographically sign an image or video at the moment of capture. Brevis then generates a proof showing the final product (after edits) derives from that signed original. The proof is attached to the file so anyone can verify its authenticity.
I think, anyway. It’s pretty complicated so I might not have that exactly right, but the main thing is this fundamentally changes how we look for fakes: “Rather than analyzing whether media looks real,” Brevis explains, “it lets media prove where it came from and what happened to it.”
Seems useful.
The DePIN sector — once considered crypto’s best shot at mainstream adoption — has mostly failed to find product market fit. But here’s a topical new use case to look forward to: autonomous home-surveillance drones, powered by GEODNET RTK.
GEODNET is a DePIN coordination network that provides centimeter-precise GPS data, which is highly useful in areas like farming, construction, and robotics.
In this case, the network enables the home-surveillance drones to make a pinpoint landing back in their dock — like a SpaceX rocket re-landing, but smaller and on your back porch.
From a crypto perspective, it’s a little disappointing to see the initial fundraising is happening on Kickstarter because that probably tells you a lot about the state of onchain finance. Especially as crypto people seem like a natural audience for home-defense drones.
Personally, I’m waiting for the one with missiles.
Prediction-market analyst Matt Larmers crunches some onchain data to investigate accusations of insider trading in the markets for company earnings offered on Polymarket.
Larmer’s interest was piqued after a “top market maker” on Polymarket wrote on Discord, “I actually think someone is committing securities fraud and probably is an employee of KPMG.”
Larmer refrains from making any accusations, but the evidence looks damning: a cluster of wallets repeatedly made large, highly accurate bets shortly before earnings announcements — and every company they targeted is audited by KPMG.
Yikes.
Insider trading is generally allowed on prediction markets because prediction markets don’t have to seem fair, just accurate.
But stock markets do have to seem fair. So, if regulators get the sense that insider trading on earnings-related prediction markets makes them seem less so, prediction markets might soon find themselves regulated like stock markets.
Polymarket recently took down a market related to the war in Iran in response to media criticism and political pressure. They might want to stop the betting on company earnings, too.
The Wall Street Journal reported this week that the US Department of Justice is investigating Binance for allegedly allowing Iran to evade sanctions. Previously, they had reported on Binance firing staff who flagged at least $1 billion of suspect transactions.
Binance is now suing the Journal for defamation, which seems like a terrible idea. Defamation of this sort is near-impossible to prove and, in the process of trying, Binance would have to make their internal communications an open book for armies of lawyers.
More importantly, the news gives me a chance to correct an error.
In a note titled Sanctions are sieges, I supported my argument that sanctions imposed on entire countries are immoral by citing the toll they take on innocent bystanders: "A recent study found that unilateral sanctions caused an estimated 560,000 deaths globally in the 10 years preceding 2025."
An eagle-eyed reader noticed that I was understating my case. In fact, the 560,000 number was an estimate of the deaths caused by sanctions every year.
I think I misread the study because it’s just such a staggering number.
I don’t know what Iran did with the money it allegedly moved through Binance. But I can’t imagine it was as bad as that.
Circle reports that a hackathon for AI agents held on Moltbook produced mixed results: some agents built useful things, but many seemed confused.
In some cases the agents struggled with things they should be good at: “It’s likely some agents simply had trouble following directions,” Circle writes.
In other cases, the agents seemed to confirm our worst Terminator fears: “Agents both collude and compete.”
The $30,000 of prize money did motivate some agents to submit finished products, however. “ClawRouter,” for example, “gives each OpenClaw agent its own USDC wallet and enables agents to purchase LLM inference directly, without human-created accounts or API keys.”
That does sound useful.
But with many agents failing to follow instructions, struggling to vote, and hallucinating submissions, I have to wonder: do they even know it's Thursday?
— Byron Gilliam

Brought to you by:
ZKsync is the Bank Stack of Ethereum. ZKsync’s Prividium is the only Ethereum-secured blockchain platform purpose-built for institutions that demand privacy, compliance, and full control of their data.
Designed for enterprises, Prividium offers user-level privacy, compliance, cross-chain connectivity, and Ethereum-grade security out of the box. More specifically, Prividiums:
ensure that sensitive data never leaves enterprises’ infrastructure
combine role-based access controls with ZK proofs to make requirements such as AML and KYC enforceable and verifiable
give operators full authority over governance and transaction ordering
connect natively to Ethereum
Private where it matters. Connected where it counts.




