BlackHart
Go live
Hacks feed/Polymarket

Polymarket

October 29, 2026·Polygon·Other
$2.7M
total loss
StatusConfirmed
View the Polymarket score →

Why it happened

Attack class: Supply chain / dependency

A hijacked monitoring script acted with users' open sessions, selling positions and moving funds through the gasless relayer.

What would have stopped it

  • Pin third-party scripts to reviewed versions and check their integrity before they run.
  • Sandbox monitoring and analytics tags away from the logged-in trading session.

What the score saw

646Forged, scored May 22, 2026

Our assessment of Polymarket was not rescored after the June vendor attack, and its supply-chain review still looked only at smart-contract dependencies, so third-party website scripts stayed outside what we measured. It did rate operational security as the weakest area, which is where front-end script controls belong.

Score dimensions

  • Supply chainmissed
  • Operational securitypartly flagged
  • Adversarial resiliencepartly flagged

Attackers took over the domain records of DebugBear, a website monitoring vendor whose script ran on Polymarket's international site, and used it to serve malicious code. For about an hour on October 29, users who had the site open had positions sold and funds moved without their consent, about 2.7 million dollars from 726 wallets according to Polymarket. Polymarket's contracts were not exploited, and it has returned 3.9 million dollars to traders, including payments for losses during the trading pause. The stolen funds were converted to POL on Polygon, and most of it still sits in three attacker wallets, with part bridged to Ethereum.

Smart contractssafe
Users without an open sessionsafe
Polymarket US exchangesafe
Open-session walletsdrained(about 2.7 million dollars from 726 wallets, refunded)
Traders hit by the pausepartially affected(about 1.19 million dollars compensated)
Exploit anatomy

A hijacked monitoring script on Polymarket's site used open sessions to move victims' funds through Polymarket's relayer, starting with a first theft into the attacker wallet, which gathered about 28.6 million POL and split it into three wallets, the largest being this one.

The attack in 3 steps. Step 1, Root control: Victim deposit wallets (source). Step 2, Relayed theft: Polymarket Exploiter 1 (attacker), about $2.64M reached this step. Step 3, Dispersal: Split wallet A (exit), Split wallet B (exit), Split wallet C (exit), about $1.72M reached this step.

  1. Step 1: Root control

    • Source
      Victim deposit wallets
      726 wallets with an open session
      ~ $2.7M
  2. Step 2: Relayed theft

  3. Step 3: Dispersal

Why each step worked

    • Vendor domain hijackedDebugBear DNS records changed through its registrar
    • Third-party tag with session accessmonitoring script ran inside the logged-in page
    • Relayer accepts session requests0x00000000000Fb5C9ADea0298D729A0CB3823Cc07
    • No step-up check on large outflowsmissing controlpositions sold and funds moved without a fresh confirmation
    • No script integrity controlmissing controlno pinned version or integrity check on the vendor file
Full forensic detail

Step-by-step reconstruction, root cause, counterfactuals, remediation, and disclosure timeline.

Exploit anatomy

1.
A performance-monitoring script from the vendor DebugBear, embedded on Polymarket's international site, began serving malicious code. DebugBear later said its DNS records had been compromised through its domain registrar, so the attacker could serve a file of their choosing from the vendor's own hostnames.
2.
For users with an active session, the script sold their positions and moved funds out of their wallets, per Polymarket's postmortem. The requests went through Polymarket's gasless relayer like normal user actions. The first theft, at 17:10 UTC, converted a victim's 181.83 pUSD into 1,957 POL and sent it to the attacker wallet 0x3F99139F1C5854bb628c2e576Aaf638e99dFfc18.
3.
Some victims' outcome tokens were transferred to the attacker instead, starting at 17:26 UTC. By 18:03 UTC the attacker wallet held about 28.6 million POL, about 2.64 million dollars, collected from hundreds of accounts.
0x4D97DCd9...476045safeBatchTransferFrom(address,address,uint256[],uint256[],bytes)tx:0xf77f91ae...eb5e99
4.
Polymarket shipped a build without the vendor script at 17:59 UTC, then set the exchange to cancel-only and paused the relayer, which ended the drain.
5.
Between 18:28 and 18:38 UTC the attacker split the POL into three wallets: 18.6 million to 0xe4C4C39C1733e1C7c06B79f97ED442B68Df41EC6, 4.3 million to 0x5392587814Ce8c6af6789764f88e5e9AB0b9962C and 5.7 million to 0xc91ee0574a14446DE200923729c349B9CAE19b60. Part was later bridged to Ethereum; about 23.9 million POL still sits in those three wallets.

Root cause

The root cause was, for the second time in four months, outside code running inside Polymarket's website. Polymarket embedded a real user monitoring script from DebugBear, a performance vendor. DebugBear says its DNS records were compromised through its domain registrar, which let the attacker serve their own file from the vendor's hostnames. Polymarket's page loaded it with full access to the user's logged-in session. For users who had the site open, the code sold positions and moved funds, and Polymarket's gasless relayer executed those requests through the Deposit Wallet Factory as if the users had made them. No smart contract was exploited and no private key was stolen. What failed was the trust boundary of the front end: a third-party tag could act with the user's session, there was no pinned version or integrity check on the file, and large outflows needed no extra confirmation. Polymarket's own figure is about 2.7 million dollars from 726 wallets; Bitquery counted 737 accounts and 985 withdrawals in about an hour. Polymarket reports it has returned 3.9 million dollars in total, including 1.19 million dollars for losses during the trading pause.

Prevention analysis

Pin third-party scripts to reviewed versions and check their integrity before they run.

A file swapped through a hijacked vendor domain would have failed the integrity check and never executed, whatever the vendor's DNS said.

Sandbox monitoring and analytics tags away from the logged-in trading session.

A performance script has no need to reach a user's session or trading client. Isolating it would have left the malicious version able to measure page speed and nothing else.

Require a fresh confirmation for large sells and withdrawals from a session.

A step-up check on full-balance withdrawals or bulk position sales would have stopped silent drains from sessions that were simply open in a browser tab.

Rescore and harden after the first vendor incident.

The June attack came through the same kind of dependency. A full review of every third-party script after June would likely have removed or contained this one.

Similar incidents

Polymarket (June 2026 vendor script)

Same platform, same class four months earlier: a compromised third-party script on the official site pushed users into signing transfers, and the relayer executed them.

Across (November 2026 domain hijack)

A registrar-level compromise let an attacker control what a trusted domain served, and users who trusted the site were drained. Here it was the vendor's domain, at Across it was the protocol's own.

Ledger Connect Kit (2023)

A trusted third-party front-end library was replaced with a drainer and ran inside many dApps at once. Same lesson: every outside script on a signing page is part of the attack surface.

Remediation

1.Polymarket removed the compromised script, cleared caches, paused trading and the relayer, and reset all active login sessions.
2.Polymarket returned 3.9 million dollars to traders: about 2.7 million dollars for drained wallets and positions and 1.19 million dollars for losses during the pause, with 95 percent of affected wallets refunded in full by October 30.
3.Polymarket says it is redeploying third-party scripts with a stronger content security policy, explicit version pinning, sandboxing and continuous monitoring, and has referred the case to law enforcement.
4.Other platforms should treat every third-party tag on a logged-in or signing page as code with full account access, and pin, check and isolate it accordingly.
5.Vendors serving scripts to financial sites should lock registrar accounts and watch DNS changes for their script and data hostnames.

Timeline

2026-10-27Last-modified time on the malicious file served from the vendor's CDN, per DeFiPrime; it may not mark when the code went live.
2026-10-29First theft: a victim's pUSD is converted to 1,957 POL and sent to the attacker through Polymarket's relayer.
2026-10-29First batch of victims' outcome tokens arrives at the attacker wallet.
2026-10-29Polymarket identifies unauthorized activity affecting user wallets.
2026-10-29Polymarket ships a build that removes the vendor's monitoring script.
2026-10-29DebugBear says it is affected by a security incident and asks monitoring customers to remove its embed code.
2026-10-29Last incoming transfer; the attacker wallet holds about 28.5 million POL.
2026-10-29Polymarket marks the incident resolved after pausing trading and the relayer.
2026-10-29Attacker begins splitting the POL into three wallets.
2026-10-30Polymarket says 726 wallets were affected and all but 37 have been made whole.
2026-10-30Polymarket publishes its postmortem: 3.9 million dollars returned in total.
2026-10-30DebugBear says its DNS records were compromised through its domain registrar.
Polymarket (containment and postmortem), DebugBear (vendor disclosure), Hudson Rock (script analysis), Bitquery (withdrawal count), DeFiPrime (on-chain timeline), trader reports on X (first public alarm), BlackHart (on-chain trace)
Continuous adversarial monitoring

Get your protocol scored across 12 dimensions, or request ongoing coverage.