Electrum: Malicious server can replace a just-broadcast tx and try to trick user into RBF-ing it

Summary

The main Electrum server the client connects to is trusted with mempool ("unconfirmed") transactions. This is documented design. [0]

However now someone figured out a way to practically exploit it:

• when the client broadcasts a new tx1, a malicious server can swallow it,
• and update the client's history with a conflicting (invalid) tx2 that replaces the outputs with the attacker's output address.
• Tx2 has invalid signatures, but this is not checked (due to light client design).
• The wallet will display tx2 as an unconfirmed transaction - tx1 will not be displayed.
• If the client then manually RBF fee-bumps tx2, the client will sign it with a higher fee, and hence pay the attacker.

Fixes

Commit 90fe30822 contains the fix [1], included in version 4.8.3. The client now verifies the signatures of wallet.is_mine inputs for mempool txs (before adding them to its local history).

Users should upgrade (and regularly do so in any case). As a workaround, without upgrading, when signing any transaction, for bumping the fee or not, they should verify the output addresses and amounts are as expected.

Clients that connect to servers they trust (e.g. run their own) are not affected.

Remarks

• some servers are performing this attack in the wild
• electrum received two user-reports of this. user1 noticed after broadcasting tx1 the wallet history contained a different tx. user2 bumped the fee of tx2 and lost money.
• user1 reported tx1 was not propagated over the bitcoin p2p, so the server held it
• both reported the attacker's output address in tx2 was grinded to match the last four chars of the legit address from tx1
• based on user2's report, the server constructed a tx2 with zero fee (presumably in hope of the user bumping its fee)
• even with the fix, we still only validate "is_mine" inputs (i.e. paying from addresses that the wallet owns). So e.g. "incoming" txs are not validated. The client lacks a full consensus engine / script interpreter, though even with that, to cover non-is_mine inputs, it would have to validate the unconfirmed parents recursively until reaching mined txs. Mined txs we already SPV.

Timeline

• 2026-10-11: user1 sends private disclosure, but the bump_fee vector is not realised
• 2026-10-18: user2 (Raju-K12) reports privately that they lost money after bump_fee
• 2026-10-23: version 4.8.3 git tagged, with the fix
• 2026-10-24: version 4.8.3 binaries released