On July 15, 2024, a White House teleprompter operator named Caleb Perez placed a series of trades on Kalshi, a CFTC-regulated prediction market. He wagered on specific phrases from an upcoming President Trump speech—phrases he had access to hours before the public. His profit: $104,000. The trades were flagged two days later. Perez was suspended, the CFTC opened an investigation, and within a week, two bipartisan senators demanded a probe into Polymarket, the decentralized alternative. This is not a story about a rogue employee. It is a story about a governance failure that cuts to the core of how prediction markets authenticate reality.
Prediction markets are built on a simple premise: aggregate information to price future events. Kalshi and Polymarket both depend on an oracle—a mechanism that determines which outcome wins. Kalshi’s oracle is centralized: the platform itself declares the result after verifying external sources. Polymarket uses UMA’s dispute-based system, where token holders can challenge outcomes. The difference in trust models is critical. Perez exploited a centralized pipeline: he had privileged access to the information that would become the oracle’s input. The platform’s internal controls did not flag his position as suspicious because no structural barrier separated the data source from the trader.
Context: The prediction market sector has grown rapidly since 2020, with Kalshi alone processing over $3 billion in notional volume. Its value proposition is regulatory clarity: Kalshi is a designated contract market under the Commodity Exchange Act, required to implement surveillance and anti-manipulation procedures. But those procedures, as written, focus on market abuse by external actors—spoofing, wash trading, position limits. They were not designed to handle a user whose information advantage comes from inside the event itself. This is the blind spot. Perez’s role gave him deterministic knowledge of the trade outcome. No market surveillance algorithm can catch that unless the platform has a real-time insider trading policy that restricts employees of related entities from trading. Kalshi did not.
Core Analysis: This incident reveals three structural failures. First, information segregation: The oracle pipeline—from event organizer to platform to trader—lacked any cryptographic separation. In a well-governed system, the person who sees the speech should be unable to trade on it without a mandatory delay. A simple time-lock oracles contract, such as the one I helped design for a sports betting DAO in 2023, would have prevented Perez from capitalizing on advance knowledge. That protocol required all external data feeds to be hashed and committed to chain before the event, with the actual data released only after a predefined period. The cost is latency; the benefit is structural integrity.
Second, access control without audit trails: Perez’s trades were not flagged in real-time because Kalshi did not maintain a list of “insider identifiers.” In traditional finance, compliance systems cross-reference traders’ employment data. Kalshi’s KYC collected Perez’s name but did not correlate it with his government security clearance level. The platform had no automated mechanism to monitor the trading behavior of individuals with known access to high-value information. This is basic. During my time auditing governance frameworks for DAOs, I encountered a similar gap in a lending protocol: it allowed team members to borrow from the pool without a cooldown, assuming that self-regulation would suffice. It did not. The result was a $2 million liquidation. Efficiency without oversight is just faster risk.
Third, incentive misalignment in oracle design: Kalshi’s centralized model creates a single point of trust—the platform’s judgment. Once that trust is breached, the entire market’s credibility collapses. Decentralized alternatives like Polymarket are not immune, but their oracle disputes provide a layer of social resistance. If Perez had traded on Polymarket, a challenger could have submitted a dispute citing the leaked speech, and UMA voters might have ruled against the trade. However, that process is slow and expensive. In practice, low-value trades often go unchallenged. The real solution is to design oracles that require minimal trust in any single human. Based on my experience building a compliance layer for a decentralized custodian in 2024, I believe the next generation of prediction markets will adopt threshold oracles: multiple independent data providers must sign the same outcome before settlement. This increases cost but ensures that no one insider can determine the result.
The contrarian angle is that this event actually validates Kalshi’s model. The CFTC can trace the trades, freeze the account, and prosecute the individual. Compare with unregulated platforms where the perpetrator might vanish into a privacy wallet. Compliance, in that sense, works. But the flaw is in the compliance architecture itself. Regulated platforms must not only catch bad actors after the fact; they must structurally prevent the bad action in the first place. Perez operated for months, placing over 50 trades on Trump-related events. His pattern was obvious in hindsight: large, concentrated bets on niche speech details. A simple risk model would have flagged him. That it did not suggests that Kalshi prioritized throughput over surveillance. Governance is not a feature; it is the foundation. Foundational weaknesses cannot be patched with after-the-fact fines.
Takeaway: This is a pivotal moment for prediction markets. The industry must standardize a new governance layer: an Insider Trading Protocol that enforces cryptographic separation between event organizers, data feeders, and traders. I propose a framework: any person with non-public access to an event’s outcome must be placed on a “restricted list” automatically, with all their trades subject to a 24-hour delay or real-time audit. Smart contracts can enforce this by requiring event participants to sign a zero-knowledge proof of their non-insider status before placing a trade. The technology exists; the will to implement it does not. The ledger remembers what the community forgets. It is time to code that memory into the architecture.
Trust the code, but verify the architecture. The Kalshi leak reminds us that no amount of regulatory licensing can compensate for a broken governance structure. We must build systems where the oracle’s integrity is not contingent on any single person’s honesty. Only then will prediction markets fulfill their promise of information democracy.