150 Wallets Turned US Military Secrets Into $8 Million

A pattern that only shows up when you isolate the right bets

The Anti-Corruption Data Collective, a nonprofit research group, did not scan all of Polymarket's activity looking for fraud. It isolated a specific shape of trade: wagers of at least $2,500 placed within a single hour on an outcome the market priced at 35 percent or lower, the kind of bet an ordinary trader loses on most of the time. Reuters reported on 20 August 2026 that this filter surfaced 556 wallets the researchers nicknamed "Orcas" for a recurring pattern - appear, win big on a long-shot, then vanish.

Of those, 152 wallets concentrated specifically in military and defense markets, together earning approximately $8 million with a 97.2 percent win rate on those bets, against a much lower baseline success rate for long-shot wagers across the platform generally. Before the June 2025 US strikes on Iranian nuclear facilities, an Orca wager on military action was followed within minutes by copycat bets of $200,000 from an automated trading bot and $100,000 from another large trader - the kind of signal cascade the researchers say makes the platform a broadcast mechanism for whoever holds the original non-public information.

This is not a first offense, and prosecutors already have a playbook

The insider-trading question on Polymarket is not new, which is what makes this data notable rather than novel. A US Army servicemember was charged by the Commodity Futures Trading Commission with using classified information about a Venezuela operation, purchasing 436,000 "Yes" shares and generating $404,000 in profit; he has pleaded not guilty. A Google software engineer was separately charged with using confidential internal Year in Search data to trade 23 Polymarket contracts with near-perfect accuracy, earning about $1.2 million. CFTC Chairman Michael Selig said in April the agency would investigate fraud, manipulation and insider trading on prediction markets.

Polymarket told Reuters that it closely monitors suspicious activity and has referred dozens of wallets to authorities, but the company did not respond to Reuters' questions about this specific research. That non-answer is itself informative: a platform that already refers cases it catches is implicitly conceding it does not catch all of them, and 152 wallets with a 97.2 percent win rate on the exact category of bet most likely to carry non-public information is a wide margin to have missed.

Why this is a European regulatory question before it is an American one

The obvious reading of this story is an American one: a US platform, US military information, US prosecutions. The more useful reading for a European reader is forward-looking. Polymarket has been actively pursuing a regulated footprint outside the US, and prediction markets more broadly are the subject of live licensing and gambling-classification debates in UK and EU regulatory circles. Every one of those conversations currently treats "can the platform police itself" as an open question. This research answers it: even with in-house monitoring and voluntary referrals, a sophisticated wallet-tracing effort by outside researchers found a concentration of near-perfect, oversized, minute-precise bets that internal systems either missed or did not act on before Reuters asked.

The following table lays out the two known coefficient categories this pattern falls into and the totals the research group attached to each.

The researchers' own recommendation is blunt: require identity verification for all traders, withhold payouts on flagged trades pending investigation, and ban markets where non-public information carries the most exploitable value. Whichever UK or EU body eventually licenses a prediction-market operator will be deciding, in effect, whether to accept the platform's self-policing record as sufficient, or to write the outside verification into the license itself.

Read next: A Federal Licence Did Not Stop New York  |  Klarna Cut Guidance the Day Its CFO and CMO Left