556 wallets, 78,496 bets: Polymarket's insider problem now has a denominator

The Anti-Corruption Data Collective has published "Classifying Insider Trading Risk," a report that examined 78,496 longshot bets placed by 12,355 wallets on Polymarket and classified 556 of those wallets as likely insider traders ^[1][2]. The count matters less than the denominator: insider trading on prediction markets has so far been handled one defendant at a time, and a population estimate is the form in which an anecdote becomes a compliance exposure.

ACDC's screen is mechanical. It defines a longshot as any wager above $2,500 at a price of 35 cents or less, meaning the market gave the outcome a 35% chance or worse ^[3]. Sorting the wallets that placed such bets produces four buckets: 556 "Orcas" that concentrate in a handful of markets and topics yet win more than 75% of the time ^[4], 3,278 "Whales" trading high volume across many markets ^[5], 760 "Bots" that appear at least partly automated ^[6], and 7,761 "Small Fish" ^[7]. The four groups account for the entire sample ^[1], which puts the Orcas at roughly 4.5% of longshot bettors ^[2].

The base rates are where the anomaly sits. In an earlier April report, ACDC found that 14% of longshot bets win across Polymarket as a whole, 25% win in political markets, and 52% win in military and defense markets ^[8] - about 3.7 times the platform-wide rate ^[3]. Report authors Michelle Kendler-Kretsch and David Szakonyi count $9.3 million in winning longshot bets from military markets alone ^[9].

The sequencing finding is the more consequential one for the venue. According to ACDC, Orcas bet first and Whales and Bots follow ^[10]. That copycat trading is legal, but it means the public ledger rebroadcasts whatever the informed wallet appeared to know ^[11]. The report's example: an Orca bet on US military action in Iran hours before the June 2025 strikes, trailed by a $200,000 bot bet and a $100,000 whale bet ^[12]. ACDC's argument is that the same transparency is available to foreign militaries and intelligence services ^[13]. Enforcement runs the other way: the money trail is pseudonymous, and exchanges holding funds for thousands of customers would have to be subpoenaed ^[14]. That is why ACDC concludes that banning the highest-risk categories outright is the only workable remedy ^[15].

ACDC is an advocacy outfit and its recommendation follows its framing. The prosecutions are separate. In April the Justice Department charged US Army soldier Gannon Ken Van Dyke with staking about $33,034 on Venezuela contracts using classified details of the raid that captured Nicolas Maduro, then collecting about $409,881 ^[16], roughly 12 times the stake ^[4]. In May, prosecutors in the Southern District of New York charged Google engineer Michele Spagnuolo with using internal company data to win $1.2 million on Polymarket ^[17]. Bubblemaps told CBS's "60 Minutes" it had traced nine linked accounts that won more than $2.4 million across more than 80 Iran war bets at a 98% win rate ^[18]. "Luck alone cannot explain those numbers," Bubblemaps chief executive Nicolas Vaiman said ^[19].

Watch whether Polymarket restricts or delists military and defense contracts, since that is where the win-rate gap is concentrated ^[8]. Watch whether any prosecution attaches a name to one of the 556 flagged wallets, which would move the classification from statistical inference to evidence ^[2][14]. And watch the copycat layer, because whales and bots shadowing an informed wallet are the mechanism that turns a single leak into a public signal ^[10][11].