Hook
Over the past seven days, a single event has reshaped how institutional observers view the intersection of regulatory compliance and prediction market integrity. On March 12, 2025, ABC News broke the story that James Perez, a 27-year-old White House teleprompter operator, had made over $100,000 in profits on Kalshi by betting on specific words and phrases that would appear in President Biden's public addresses—including the State of the Union. He did this not by analyzing policy shifts, but by reading the actual speech script days before delivery. The CFTC is currently in settlement negotiations with Perez. This is not a random anomaly; it is the first documented case of insider trading in a CFTC-regulated prediction market. And it reveals a structural weakness that no smart contract audit could have caught: the human information advantage.
Context
Kalshi is a U.S.-based prediction market platform operating under CFTC oversight as a designated contract market (DCM). Unlike its decentralized counterpart Polymarket, Kalshi requires KYC/AML, mandates employer disclosure, and maintains a dedicated surveillance team that flags suspicious trading patterns. The platform lists event contracts on political outcomes, economic indicators, and even specific word mentions in speeches. These “Mentions” markets—where users bet on whether a particular word appears in a president’s speech—have existed for years. They attract speculators, but also those who might possess non-public information. Kalshi’s rules explicitly prohibit trading based on material non-public information, and since last month, the platform has strengthened its employer verification process. Yet, Perez still managed to place 14 separate bets over multiple speeches, including the State of the Union, and even adjusted his positions mid-speech as he fed the teleprompter text to the president. The platform’s surveillance team flagged his activity and reported it to the CFTC. This is not a story about code failure. It is a story about information asymmetry and the limits of platform-level monitoring.
Core
Let’s dissect the technical and procedural elements that enabled this trade. First, the product design: the “Mentions” market is inherently vulnerable to information asymmetry. Unlike a binary market on “Will the Fed raise rates?” which relies on widely distributed data, a market on “Will the president say the word ‘infrastructure’?” is a direct function of a single document—the speech script—held by a few dozen people. From a cryptographic perspective, this is a classic “secret knowledge” problem. The platform cannot cryptographically enforce that no trader has access to the speech text because the text itself is not on-chain. Kalshi’s defense relies on legal prohibitions and reactive monitoring, not proactive technological prevention. Based on my audit experience in 2022, when I analyzed the Oracle integration failures of 12 DeFi protocols, I learned that any system relying on off-chain information that is not publicly verifiable before the event is vulnerable to a “time-of-knowledge” attack. Perez had knowledge hours or days before the speech was delivered. That’s an information advantage that no consensus mechanism can mitigate.
Second, the monitoring system. Kalshi’s surveillance team flagged Perez because his trading pattern was anomalous: repeated high-conviction bets on the exact words that later appeared, with concentration in time around speech events. But flagging is not blocking. The trades still executed. The platform’s architecture is fundamentally reactive—it relies on post-hoc analysis of trading data. Compare this to traditional exchange surveillance systems, which can freeze accounts preemptively if an insider trading flag is triggered. Kalshi’s system appears to lack real-time execution halts. This is a critical design flaw. In my 2024 deep dive into BlackRock’s BUIDL settlement layers, I observed that institutional-grade compliance systems implement pre-trade validation for KYC/AML attributes, not post-trade alerts. Kalshi’s current setup works for after-the-fact reporting, but it does not prevent the crime. The 2020 Compound Finance stress test I conducted taught me that conservative design must anticipate worst-case behavior. Here, the worst case is an insider with direct access to the outcome document. The system did not stop him.
Third, the regulatory loophole. The CFTC’s jurisdiction over prediction market insider trading is nascent. The agency has not yet formalized a definition of “non-public information” specific to event contracts. In securities law, insider trading is well-defined: trading on material, non-public information in breach of a duty of trust. In the prediction market context, what constitutes a “duty”? Perez, as a government employee, had a duty to safeguard the speech text. But does a private contractor building the teleprompter software have the same duty? Does a speechwriter’s roommate? The ambiguity creates enforcement gaps. The fact that the FBI has only investigated two prior prediction market insider trading cases—one involving Venezuelan President Maduro’s speech and another concerning a Google employee—shows how underdeveloped the legal framework is. This case will likely become a template for future rulings.
Contrarian
Most commentary focuses on Kalshi’s failure to prevent insider trading. But the contrarian angle is that Kalshi’s active reporting is actually a net positive for the platform’s long-term credibility—and for the prediction market industry as a whole. By flagging Perez, Kalshi demonstrated that its surveillance system works at the detection level. This is more than Polymarket—a decentralized platform without KYC—can realistically achieve. In Palomarket, insider trading is structurally undetectable because the underlying identity is pseudonymous. A similar trade could be executed by a pseudonymous address, and there is no central authority to report to regulators. So Kalshi’s “transparent compliance” narrative may actually strengthen its case for regulatory approval of new contract types. The CFTC needs data points to build its rules. Kalshi provides that data willingly. This is not a security weakness; it’s a PR asset disguised as a scandal.
However, the blind spot is deeper. The presumption that “users will disclose their employer honestly” is fundamentally naive. Perez disclosed his White House employment. Kalshi knew he worked there. The platform still allowed him to trade in speech-related markets. The question is: why didn’t the system automatically restrict White House employees from trading in “Presidential Mentions” markets? This is a basic conflict-of-interest filter that any competent compliance engineer would implement. In my 2017 audit of the Golem token sale, I found integer overflows that the developers had missed because they assumed inputs would be reasonable. Here, the assumption is that employer information is solely for KYC—not for dynamic trading restrictions. This is a design oversight that must be addressed immediately.
Another blind spot: the scalability of monitoring. Kalshi currently flags manually. As trading volume grows, they will need automated pattern recognition models. But those models can be gamed. An insider could spread their bets across multiple accounts, spread the timing, or use correlated markets to hide intent. The cat-and-mouse game of surveillance is well-known in traditional finance. In the crypto world, where on-chain data is transparent, we can at least trace transactions. Here, Kalshi operates on a centralized ledger, so external auditors cannot verify the surveillance integrity. The system is centralized and opaque by design.
Takeaway
The Perez case is not a crisis—it’s a calibration point. Kalshi’s compliance team passed the test of detection, but the architecture failed at prevention. The next iteration must include pre-trade filtering based on employer codes and contract sensitivity. The CFTC will likely require this as part of the settlement. For the prediction market sector, this event accelerates the regulatory divide: compliant platforms like Kalshi will invest in heavy surveillance infrastructure, while decentralized platforms will remain leaky but permissionless. Long term, the winner will be the one that can cryptographically prove no insider knowledge was used—not just promise to monitor. Trust no one, verify the proof, sign the block. For now, the proof is incomplete. The block is still pending.
