Polymarket exploit analysis

Polymarket, a prediction-markets venue built on Polygon, lost roughly $2.94 million on 25 June 2026 in what security researchers classified as a frontend supply-chain attack rather than any flaw in the protocol's on-chain contracts. The web3isgoinggreat timeline entry on the incident, citing Polymarket's own announcement and the on-chain investigator Specter, places the loss at about $2.97 million; Crypto.news and Cointelegraph reporting the next day settled on $2.94 million; the consensus range across coverage is $2.9 to $3 million.

The breach sat in a third-party vendor used by the Polymarket website frontend, not in Polymarket's trading contracts or in its wallet integration. The attacker used the vendor's access to inject a malicious script into pages served to logged-in Polymarket users. The script phished for wallet signatures that authorised the transfer of pUSD positions; once signed, the attacker drained those balances into attacker-controlled wallets and swapped the proceeds into ETH.

Polymarket confirmed the pattern in a public statement and committed to refunding affected users out of treasury funds. AMLBot, cited by Protos and Cointelegraph, attributed the on-chain portion of the theft to 11 wallets and traced the proceeds to Ethereum. Several outlets noted that this was Polymarket's second publicly disclosed breach within two months; the previous incident is not the subject of this article.

The interesting detail is the placement of the vulnerability. Prediction-market venues typically hold user funds in self-custodial wallet connectors, so a frontend compromise by itself cannot move those funds unless the user actively signs a malicious transaction prompt. CryptoPotato's framing of the incident as a supply-chain attack is accurate; the rest of the threat is the user signing what looks like a routine confirmation but is in fact an authorisation to move pUSD.

The Polymarket incident sits inside a growing class of frontend-side compromises at venues where users are expected to self-custody. The on-chain threat model holds up; the off-chain surface area does not. Watching what the frontend does, rather than what the contracts say, is the layer that matters here.

This analysis was prepared by reading the web3isgoinggreat timeline entry on the incident and the meta descriptions of follow-up reporting from Crypto.news and CryptoPotato, the only published coverage of this specific incident that fetched successfully.