NEWS
Polymarket Halted Trading After a Page-Speed Script Drained Wallets
A DebugBear page-speed script, not a broken contract, emptied 737 Polymarket accounts of $2.73 million on October 29. Trading resumed within an hour.
A hijacked page-speed script emptied $2.73 million from 737 Polymarket accounts on October 29, then the exchange halted trading for one hour. Bitquery’s on-chain review found every drain was a normal withdrawal through Polymarket’s own relayer. No smart contract broke.
The file came from DebugBear, a website performance firm Polymarket still has not named. Logged-in visitors on polymarket.com were the ones in the blast radius, including people who never placed a trade that hour.
The Vendor Polymarket Still Has Not Named
Polymarket’s public line was a “3rd party provider.” Its production bundle was more specific. Build metadata timestamped 17:59:20 UTC carried the change message “chore(polymarket): remove debugbear rum script (#15337).”
DebugBear sells lab tests and real user monitoring in the browser. A RUM embed runs on the live page so the vendor can time how fast that page feels. On a news site that is a timing beacon. On Polymarket it sat in the same tab as the order book and the signer.
At 18:00:58 UTC, 98 seconds after that build stamp, DebugBear told every RUM customer to pull the embed, not just one client.
We’re impacted by an ongoing security incident. If you use our RUM product, please remove the embed code from your website.
DebugBear, on X, October 29, 2026
Hudson Rock later downloaded cdn.debugbear.com/C7xcKKrjUnXK.js twice, at 18:12:06 and 18:13:04 UTC, and got the same file both times. Nothing in those 15,187 bytes measured paint times. The logic was written against Polymarket’s login keys, React tree and relayer. Polymarket referred the incident to law enforcement and promised to make affected users whole. It has not confirmed DebugBear by name.
How a Speed Script Emptied Polymarket Accounts
The contracts did what they were built to do. The hostile code used a session the user had already opened on the real site. Hudson Rock’s review of the 15,187-byte wallet-draining payload found no routine that sent a seed phrase or a private key off the machine. It reused authority that was already in the tab.
WHAT THE FILE DID ON LOAD
- Local storage: It read
polymarket.auth.params,polymarket.auth.proxyWalletandpoly_clob_api_key_map, which tell the app how the user signed in, which deposit wallet to use and which order-book keys belong to the session. - Trading client: It walked React’s internal tree, looking for an object with
postOrderandcreateOrderorcreateMarketOrder, then tried to trade as that user. - Floors and geoblock: Fallback settings skipped portfolios and positions under $10 and quit if Polymarket’s geoblock call marked the visitor as blocked.
- Telemetry cover: Portfolio summaries and transfer results went to DebugBear hostnames under paths such as
/metrics/submit, so the traffic looked like ordinary RUM.
If the live config call failed, the file fell back to a hard-coded recipient, 0x3F99139F1C5854bb628c2e576Aaf638e99dFfc18, and preferred POL as the payout token. On-chain, the drains look like user-approved batches, because they were. Approvals Bitquery sampled were used three to seven seconds after they were created. People do not click through hundreds of accounts that fast. Software does.
The File Looked for a Live Trading Client
That search is the whole trick. A performance script does not need the order-book client. The browser does not isolate third-party JavaScript from first-party code on the same page, so the embed inherited the session for free. WhoGetsMyData’s scan, cited by Hudson Rock, showed Polymarket’s content security policy allowed the DebugBear host. Allowing the host is how RUM works. It is also how a swapped file becomes trusted code.
Pinning the vendor file to a known hash, or hosting it yourself, would have made a swapped copy fail to load. After a similar frontend hit in June, that control was available. It was not in place on Thursday.
Magic Accounts Never Saw a Prompt
The file then split on login type. Email users get a Magic wallet embedded in the page. During a session that wallet can sign without a browser-extension window. The script sold positions with fill-and-kill orders, then swept the cash. That path matches users who said they were drained without signing anything, including trader CarOnPolymarket, who posted at 17:45:57 UTC that $800,000 was gone and, ten minutes later, “All my positions were sold.”
Users who connected an external wallet took a different path. The script built an ERC-1155 batch transfer of outcome tokens to the attacker and pushed it through Polymarket’s gasless relayer. Those users got a signature request. Screenshots of that prompt circulated as the transaction to reject. Self-custody did not make the tab safe. It only put a human in the loop.
$2.73 Million Left 737 Accounts in 60 Minutes
The first POL hit the payout address at 17:10:23 UTC in Polygon block 96,392,296. Bitquery followed the money and found 737 accounts lost $2.73 million between 17:10 and 18:10 UTC, across 985 drains. By 17:45 UTC, when Car warned people to lock down, 48% of the money was already gone.
THE OCTOBER 29 TOLL
| Measure | Figure |
|---|---|
| Money taken from accounts | $2.73 million |
| Accounts drained | 737 |
| Drain transactions | 985 |
| Window | 17:10 to 18:10 UTC |
| Largest single account | $358,369 (hewhodoesit) |
| Accounts that lost under $100 | 276 |
| Share taken from four accounts | 43% |
| Median loss | $171 |
Car’s $800,000 does not match any one account on that route. Bitquery notes the figure may span several accounts or count open bets at market price. The friend Car said lost $600,000 also has no single matching drain. Most of the haul still came from a few desks. Four accounts lost 43% of the money. Two hundred seventy-six accounts lost under $100 each.
At 18:19:58 UTC the payout address held 28,636,284 POL, about $2.64 million at DefiLlama’s price of $0.0922. That is converted proceeds sitting in the wallet, not a second loss total. Between 18:28 and 18:38 UTC the POL moved to three fresh wallets and began bridging toward Ethereum. At 19:49 UTC Bitquery still saw about $2.17 million of it on Polygon.
THE HOUR ON OCTOBER 29
- 17:10 UTC: First POL arrives at 0x3F99…fc18.
- 17:44 UTC: New Polymarket account gfg44433 is opened to cash stolen outcome shares.
- 17:45 UTC: CarOnPolymarket posts that $800,000 is gone.
- 17:59 UTC: Polymarket build removes the DebugBear RUM script, 48 minutes and 57 seconds after the first POL.
- 18:05 UTC: Exchange enters cancel-only mode and pauses the relayer (2:05 p.m. ET).
- 18:33 UTC: Polymarket says a third-party provider was breached and affected users will be made whole.
- 19:00 UTC: Trading is fully back (about 3 p.m. ET); the one-hour major outage record is marked resolved at 19:05 UTC.
- 19:11 UTC: One more account is drained for $10.99, 38 minutes after the fix statement.
Cancel-only and the relayer pause mattered as much as the code change. Open tabs still had the old page. Stopping new orders and unpaid gas closed both theft paths, at the cost of freezing honest users too. The relayer went quiet 30 seconds after the last drain of the hour, returned at 18:52 UTC, and then paid that stray $10.99.
Email Logins Never Asked for a Signature
Polymarket pays network fees so traders do not hold POL for gas. You sign an approval. The relayer posts it. On Thursday the relayer paid for the exit as well. Combined with email wallets that can sign inside the page, a script in the tab could sell and withdraw the way the product is meant to work.
That is why “just close the phishing tab” was the wrong instinct. People were on polymarket.com. The cleaning crew had keys to the trading floor, and someone used those keys while dressed as staff. Withdrawals also had no second factor. Once the session was live, the next click the user never made was enough.
Outcome tokens left 33 accounts too. Shares are hard to cash off-platform, so the attacker opened gfg44433 thirty-four minutes into the hour and started selling. At 19:49 UTC that account still held $4,655 in pUSD. Whoever opened it went through Polymarket’s own sign-up, which leaves records the company can pull.
Why MetaMask Prediction Markets Stayed Untouched
MetaMask offers Polymarket markets inside its own app. It does not load polymarket.com, and it does not use Polymarket’s SDKs. Users there were outside the RUM blast radius even while the international site was in cancel-only mode. The U.S. Polymarket venue was not hit either. The drain was a front-end problem on the international site, which is why a wallet that never opened that page did not pay.
📣 FYI: MetaMask prediction market users are unaffected by this issue.
While MetaMask has implemented Polymarket prediction markets, we do not use Polymarket's front end nor SDKs.
All Polymarket trading, however, is halted while they look to resolve any ongoing concern. https://t.co/Ye85lnviK3
— MetaMask 🦊 (@MetaMask) October 29, 2026
0xngmi, who founded DeFiLlama, wrote that seed phrases were safe and that anyone who visited the site was exposed. Logged-out visitors had nothing in local storage for the file to use. Logged-in visitors, on any page where the embed ran, did. Seeds were never the prize. The session was.
The Second Frontend Drain in Four Months
Thursday was Polymarket’s second vendor-script drain since June, and its third security hit since May. In May a leaked private key emptied a dormant operational wallet of roughly $600,000. In June a compromised third-party vendor injected a hostile script into the frontend; later wrap-ups Bitquery cited put that loss at about $2.94 million from about 11 wallets. Polymarket never named the June vendor. It pledged full refunds then as well.
THREE HITS IN SIX MONTHS
| When | What broke | Scale |
|---|---|---|
| May 2026 | Leaked private key on an operational wallet | Roughly $600,000, one wallet |
| June 2026 | Third-party script on the frontend | About $2.94 million, about 11 wallets |
| October 29, 2026 | Hijacked DebugBear RUM file | $2.73 million, 737 accounts |
June hit a handful of large accounts, about $267,000 each on those figures. October hit hundreds of accounts, about $3,700 each. Same kind of money, a much wider net. Hundreds of small drains in one hour is what bulk session abuse looks like, not a queue of people clicking a phishing prompt.
The drainer Hudson Rock read was written for Polymarket, and nothing public shows other DebugBear customers were served the same file. DebugBear still told every RUM client to remove the embed. Other sites that load that snippet had to treat Thursday as their incident too, even if their pages cannot place a trade.
Repayment Splits at $10,000
After the restart, Polymarket said trading was fully back and that it is fully reimbursing all affected traders. A Discord notice then drew a line. Users with less than $10,000 taken were told funds would be restored by October 30. Users above that line were told to sign a release. Some traders said the large-loss path also asks for identifying documents. Polymarket has not published an official loss total to sit beside Bitquery’s $2.73 million.
We’ve identified & fixed an issue impacting our international platform, Polymarket.com. One of our 3rd party providers was compromised, and we have referred this to relevant law enforcement. Anyone who has been impacted will be made whole.
Polymarket Traders, official account, on X
The status page lists the outage as resolved. The company has not named DebugBear, has not said whether other RUM customers saw a hostile file, and has not explained why a speed script ran in the same tab as the signer after June. Large-loss traders are still being asked to sign before the money comes back.
Disclaimer: This article is news reporting and analysis of a security incident on a prediction-market platform. It is informational only and is not investment, trading, tax or legal advice, and it is not a recommendation to use, avoid or withdraw from any venue. Readers who hold funds on a prediction market or in a crypto wallet should consult a qualified financial adviser and, if they may have been affected, the platform’s official support channels and a licensed attorney before signing any release or sending documents. Figures, wallet balances and repayment statuses reflect the primary records and on-chain reviews cited here as of the dates on those records and may change as reimbursements and investigations proceed.
-
BUSINESS5 months agoAramco Crash Kills 14 Two Days After Oil Restart
-
NEWS6 months agoSalah Captained Egypt to Their First World Cup Win
-
GAMING2 months agoZelda’s 40th Switch 2 Ships Without a Zelda Game
-
NEWS3 months agoApple Labor Day Deals Hide June’s Memory Price Hike
-
NEWS4 months agoEl Niño Puts Israel’s Needed Winter Rain on a Flood Clock
-
GAMING2 months agoPalworld 1.0.4 Adds Palpedia Search After the 1.0 Launch
-
NEWS4 months agoSaudi Visa Routes Split Nigerian Travelers Into Two Speeds
-
BUSINESS9 months agoBAE’s 60-Year Saudi Franchise Now Runs Through Taif
