Points of Focus
- Predict.fun users lost 397,708 USDT across 29 wallets in two attacks on Oct. 22 and 27.
- Investigators found that compromised wallet owner keys authorized the transactions, with no evidence of an exploit in the platform’s market contracts.
- Attackers moved 166.8 ETH into Tornado Cash, while Predict.fun promised reimbursement and investigated a possible third-party website compromise.
Prediction market platform Predict.fun has suffered about $398,000 in cryptocurrency losses across 29 user wallets, with investigators uncovering two attacks that occurred five days apart.
According to an Oct. 27 investigation by blockchain analytics company Bitquery, attackers drained 397,708 USDT from accounts on the BNB Chain-based platform between Oct. 22 and 27.
Initial reports identified 19 affected wallets and about $260,000 in losses. However, further onchain analysis revealed an earlier attack involving six wallets.
🚨 JUST IN: Hackers have reportedly drained $400,000 from https://t.co/Hp98luKNVg, a prediction market integrated into Binance Wallet.
The team says all affected users will be fully reimbursed.
The attack comes just days after Binance Futures listed its $PREDICT token.
— EyeWhales (@EyeWhales) October 28, 2026
More importantly, investigators found no evidence that Predict.fun’s prediction market smart contracts were exploited. Instead, all 33 identified draining transactions were authorized using the affected wallets’ cryptographic owner keys.
Predict.fun has preliminarily linked the incident to a possible compromise of its externally hosted API documentation website and promised to reimburse affected users. The precise method used to obtain wallet credentials remains unconfirmed.
How the Predict.fun attacks unfolded
Bitquery’s reconstruction shows the thefts began on Oct. 22, five days before the incident attracted public attention.
The first wave affected six wallets and resulted in $125,029 in losses. One account lost $109,000 across two transactions, making it the largest individual loss identified.
On Oct. 27, the attacker resumed activity, stealing $272,679.
During the most concentrated period, 19 wallets lost around $262,000 in just 34 minutes, averaging nearly one affected account every two minutes.
Investigators also identified a subsequent 4,900-USDt (USDT) theft after Predict.fun issued its public statement. That amount is included in the overall incident total.
The earlier activity raises questions about whether suspicious withdrawals could have been detected before the second wave.
Why wallet credentials, not smart contracts, were the target
Predict.fun allows traders to speculate on future outcomes using USDT as collateral.
Its accounts use ZeroDev’s Kernel smart wallet architecture, with authentication supported by Privy.
Unlike conventional wallets that require users to manage seed phrases directly, these smart wallets rely on owner keys that authorize transactions.
Bitquery found that all 33 unauthorized transfers passed the wallets’ normal signature checks.
None interacted with Predict.fun’s market or exchange contracts.
This suggests the attacker obtained credentials capable of authorizing withdrawals rather than exploiting the platform’s core trading infrastructure.
Investigators also identified Candide’s transaction bundling and gas sponsorship services in the transaction flow. Their involvement does not indicate those services were compromised.
The findings highlight an important security risk: Even when trading contracts operate correctly, stolen authentication credentials can still expose user funds.
Did Predict.fun’s API documentation expose private keys?
Predict.fun’s preliminary investigation focuses on its API documentation website, which was hosted by a third-party provider.
APIs allow developers to automate trading, but certain integrations require users to export sensitive wallet credentials.
If those credentials are exposed, attackers may be able to authorize transactions without accessing the user’s original device.
Bitquery found that 12 affected accounts had executed more than 1,000 trades during the preceding week, indicating substantial automated activity.
However, five accounts had completed fewer than 100 trades. One wallet that lost $53,700 had made just five trades.
This suggests the incident was not limited to high-frequency trading accounts.
Security researchers at MuseSec also highlighted risks associated with third-party websites operating within a platform’s trusted environment.
The documentation website has since been taken offline, although investigators have not conclusively established how the credentials were compromised.
How attackers laundered $398,000 through Tornado Cash
After draining the wallets, attackers moved the stolen USDT through intermediary addresses and cross-chain services, including Relay, LI.FI, and Across.
The funds were transferred to Ethereum, largely converted into Ether (ETH), and deposited into Tornado Cash.
Bitquery identified 166.8 ETH entering the privacy protocol across the two attack waves.
During the Oct. 27 incident, the first Tornado Cash deposit occurred less than 11 minutes after the main draining operation began.
Investigators also traced the subsequent 4,900-USDT theft through Railgun, another privacy-focused protocol.
These rapid transfers complicate recovery efforts because assets can cross several networks and enter privacy systems before investigators complete their initial tracing.
What happens next for Predict.fun users?
Predict.fun has promised full reimbursement but has not published a completed forensic report or detailed compensation timeline.
Several questions remain unresolved, including how the attacker obtained owner keys, whether the Oct. 22 withdrawals triggered internal alerts and whether additional credentials were exposed.
The incident also raises broader concerns about automated prediction market trading.
While APIs make sophisticated trading strategies possible, exporting wallet credentials can create security risks beyond blockchain smart contracts.
Affected users should follow verified platform instructions and replace potentially compromised credentials. Disconnecting a website alone may not invalidate a stolen private key.
The central lesson is that secure smart contracts do not guarantee secure user accounts. Predict.fun’s losses demonstrate how weaknesses in surrounding authentication and developer infrastructure can expose funds even when the underlying trading contracts remain intact.
Unlock premium content
Create a free account to continue reading AlphaClub articles and access exclusive features.
Share


