How Polymarket Insider Signals Are Fueling Risky Copy Trading Patterns

According to new research from the nonprofit Anti-Corruption Data Collective, a cluster of wallets on Polymarket International has been placing suspiciously timed "long-shot" bets on US military and defense markets — and other traders, including automated bots, appear to be following them in real time. For anyone in the copy-trading space, this is more than a crypto curiosity. It's a live demonstration of how signal leakage works on a transparent ledger, and why blindly tailing a hot wallet can be a fast track to your own drawdown.

What the "Orcas" actually did

ACDC analyzed every settled Polymarket market through May 5 and zeroed in on 556 wallets it labeled "Orcas" — accounts that opened, placed aggressive long-shot wagers (cumulative bets of at least $2,500 within an hour on outcomes priced at 35% or lower), frequently won, and often vanished after cashing out. Of those, 152 specialized in military and defense markets, collectively pocketing around $8 million with a reported 97.2% average win rate. ACDC is careful to note luck remains a possible explanation, and that not every suspected insider fits the Orca mold — Gannon Ken Van Dyke, the US soldier charged in April over alleged Polymarket trading tied to classified intelligence, built his position more slowly and wasn't among the 152.

The kicker for signal-followers: ACDC says these trades were visible enough to attract copycats. In one case, an Orca bet on US military action against Iran — placed hours before the June 2025 strikes — was followed by a bot and a large trader piling in with $200,000 and $100,000 respectively. Polymarket settles trades on a blockchain, so positions are public even when traders aren't. ACDC co-founder David Szakonyi put it plainly: unusual activity on the platform is more observable than many people assume because it's "all right there on the internet."

Why this matters to anyone running a copy strategy

If you've ever allocated capital to a top-ranked signal provider because of a perfect-looking equity curve, this story should hit close to home. The Orca pattern — high win rate, niche markets, short lifespan, fast exit — is the exact profile that looks magnetic in a leaderboard filter. ACDC's data suggests that at least some of those returns may have been driven by non-public information rather than edge. Copying that flow, knowingly or not, isn't alpha. It's just front-running the front-runner, and the second-mover rarely gets the same fill or the same payout.

The regulatory backdrop is tightening for a reason. US prosecutors have already brought the first insider-trading case tied to prediction markets with the Van Dyke charges, and Kalshi has started collecting employment information from traders seeking access to contracts vulnerable to insider flow, adding risk scoring and surveillance tools through Solidus Labs and Comply. ACDC is pushing for mandatory identity verification and the ability to withhold payouts on suspicious trades — and arguing that markets where non-public information is most actionable should be banned outright.

What to actually do with this

First, treat any prediction-market signal provider showing a 95%+ win rate over a short sample with the same suspicion you'd give a prop firm challenge "passer" selling a course. Win rate alone, without context on market type, position sizing, or holding period, is a survivorship-bait metric. Second, remember that on-chain visibility cuts both ways — the leader is visible to you and to every bot scanning the same mempool. By the time a copy signal is obvious, the edge is usually already monetized by faster players. And finally, if you're building any automated copy logic around prediction markets, bake in asymmetry filters: reject signals where the alleged edge is implausibly high relative to the market's information environment. The Orcas made their money because the copycat capital followed too late — make sure yours doesn't.