Kalshi's enforcement chief concedes the equity surveillance playbook does not port over

Robert DeNault, Head of Enforcement at Kalshi, and Daniel Taylor, an accounting professor at Wharton who directs its Forensic Analytics Lab, have published a joint argument that prediction markets have distinctive features which any surveillance system must account for, even as many principles carry over from equities ^[1][3]. The two write that as volumes rise, the industry's long-term success depends on both established players and new entrants adopting rigorous surveillance ^[2] - which is a venue's own enforcement side setting a bar in public, and therefore a preview of what examiners and counterparties will ask about.

The comparison they draw is unflattering by design. Equity market surveillance has had 50 years to mature, and its core principles now sit inside systems such as FINRA's SONAR and the SEC's ATLAS platform ^[4]. Those tools are built to detect illicit trading ahead of market-moving corporate events: when a stock jumps 40% on FDA trial results, the system pulls the accounts that traded shortly beforehand and tests whether that activity departs from their own history ^[5]. DeNault and Taylor add that two decades of SEC insider-trading cases suggest the agency is good at spotting the "YOLO" pattern, an unusually large out-of-the-money options position taken shortly before news ^[6].

The structural point is about what an insider can actually know. In equities, almost nobody holds material non-public information about the price itself; they hold information about one of many inputs to it, such as rates, earnings, guidance or executive turnover ^[7]. So monetisation is indirect, and the rational tactic is to wait until just before the information goes public, because until then the position carries unrelated risk ^[8]. The authors cite the SEC's hack-to-trade cases: obtaining an earnings release early does not reliably produce a profit ^[9].

Prediction markets collapse that structure. Contracts resolve against a named metric or a discrete event - the authors list Sweetgreen's quarterly profit margin, DoorDash's quarterly delivery volume, SpaceX's monthly launch count, and whether Pam Bondi or Tulsi Gabbard will step down ^[10]. Here the inside information can be the outcome itself, meaning whether the contract settles at $0 or $1 ^[11]. An insider who knows the margin buys the contract written on the margin, with the intervening factors stripped out ^[12], and the rational strategy is often to trade immediately on learning it ^[13].

That inversion is the operational problem. A surveillance model calibrated to detect clustering just before an announcement has much less to detect when the profitable trade happens at the moment the information is created, days or weeks earlier ^[1]. It also drags a new population into scope: people with access to a company's quarterly numbers now sit inside a prediction venue's surveillance perimeter without ever touching that company's stock ^[2].

Watch three things. Whether other venues publish comparable standards, given that DeNault and Taylor explicitly extend the obligation to new entrants ^[2]. Whether the venues can show timing models keyed to information creation rather than publication ^[1]. And whether issuers whose metrics are referenced in listed contracts ^[10] start treating those contracts as covered instruments in their own conduct policies, because the authors' own logic says the exposure exists whether or not anyone has written it down ^[2].