In one minute
Transfers from exposed Keplr social-login wallets were initially described as theft. The investigator later reported that the receiving addresses were among those Web3Auth identified as recovery wallets. The incident still needs a full accounting of losses and returns.
The lead story
Keplr: distinguish a recovery sweep from an attacker's proceeds
An investigation into the Keplr wallet incident changed after Web3Auth identified recovery addresses. Bitquery's investigation now says that addresses Web3Auth lists as recovery wallets held about $6.01 million at 15:38 UTC on November 7, after transfers from at least 44,041 social-login wallets on seven Cosmos chains and a separately counted set on Injective, and notes that it had first reported these sweeps as theft.
Why now: The correction changes the interpretation of the incident while affected users are still waiting for a clear account of their assets. Reporting the transferred value as an attacker's final haul would conflate the movement of assets with the identity of the recipient and the ultimate economic loss.
An angle worth explaining: Social login can make a wallet easier to access while adding a service provider to the recovery and authentication process. When that path is compromised, an emergency sweep may move assets away from exposed accounts before an attacker reaches them. A transfer to a recovery address can be protective, but users still need a reliable way to establish the amount attributable to them and receive it back.
Bitquery found that 86% of the wallets in its seven-chain count were swept after Keplr warned users. It also traced staked ATOM moved as share tokens without the usual unbonding wait. Those observations describe the investigator’s selected recovery-address flows, not an independently verified loss total.
A useful timeline separates four stages: the security issue, the original unauthorized activity, the protective movement claimed by the provider, and eventual return to users. The values at each stage can differ. A balance observed in recovery addresses is a point-in-time holding, not an amount already returned to customers. Nor should it be added to an earlier estimate without checking whether the same transfers are counted twice.
What remains uncertain: The final amount stolen, the amount ultimately returned and the recovery timetable remain unresolved in the material reviewed. The correction does not establish that every transfer in the incident was benign.
What could enter the conversation next
A published reconciliation connecting affected accounts, recovery balances and actual distributions would give users a basis to verify progress. An independent transaction analysis could also distinguish transfers to the addresses identified for recovery from any separate attacker-controlled destinations. Until then, the meaningful story is the changing attribution and the outstanding recovery process, not a precise final loss figure.
Sources (1)
- Bitquery's investigationbitquery.io, Publication date not specified