Web3Auth Disputes $6.01M Keplr Theft Estimate as Recovery Operations

Web3Auth is disputing a widely circulated $6.01 million estimate for the theft of funds from Keplr social-login wallets, arguing that a portion of the flagged on-chain outflows were whitehat recovery operations routed to addresses the company controls. The claim cannot yet be independently reconciled against the published data, though the company has provided the means to try.

The dispute sits on top of a confirmed security incident. A vulnerability in Web3Auth's services affected Keplr users who created wallets through Google and Apple sign-ins, with funds beginning to drain from those accounts over November 6 and 7. Web3Auth confirmed the flaw on November 6 and paused its embedded wallet service the same day.

How large was the breach?

Blockchain analytics firm Bitquery tracked more than 44,041 affected wallets in the Cosmos ecosystem and another 7,603 on Injective, a Cosmos-based chain. Keplr co-founder Josh Lee said the breach stemmed from a flaw in Web3Auth's services and affected roughly 20% of wallets created through Google and Apple logins — about one in five of those accounts.

Wallets created through conventional means showed no signs of compromise. According to Keplr, accounts set up with recovery phrases or hardware devices remained safe throughout the incident. That distinction matters for anyone weighing the operational trade-offs of custody models, and it is the clearest fact in an otherwise contested picture.

The $6.01 million figure came from tallying funds that flowed out of affected wallets during the incident window. Web3Auth's position is that the tally swept up the wrong transactions, because some movements labeled as theft were protective recoveries carried out by the company and its partners. On-chain, the two looked nearly identical. The recovery operations used the same technical routes as the initial attacks, so a rescue and a robbery produced almost the same forensic signature.

What does the recovery address list prove?

On November 8, Web3Auth published a list of verified recovery addresses covering eleven Cosmos-based networks and eight EVM-compatible chains. The company's argument is straightforward: if the disputed funds sit in those addresses, they were not stolen. The list lets users and outside analysts verify the claim against the chain itself rather than accept the company's accounting on faith.

That verification step is essential, because both sides of this dispute have incentives. Early damage estimates derived from raw outflow tallies can overstate losses, as Web3Auth argues happened here. But a company's own accounting of its incident deserves the same scrutiny, and the published addresses are the only mechanism through which outsiders can apply it. Neither figure — the $6.01 million estimate or Web3Auth's implied lower number — has been independently confirmed at the time of writing.

What did the companies tell users?

Both Web3Auth and Keplr directed users to ignore unofficial recovery offers and rely only on the companies' own communication channels. Keplr additionally advised users to stop depositing new funds into affected accounts and to move existing assets to secure wallets in the meantime. The guidance is standard incident response, but it underscores the practical dependency: when a provider manages key infrastructure, users cannot remediate on their own.

For non-specialists, the setup works as follows. A social-login wallet lets a user open a crypto account with a Gmail or Apple ID instead of recording a 12- or 24-word seed phrase, with the third-party provider handling key management behind the scenes. The security consequence is structural, not incidental. When users hold their own recovery phrase or hardware device, exposure depends largely on their own practices. When a provider manages the keys, a single flaw upstream can reach tens of thousands of wallets simultaneously.

What remains unresolved?

The episode also exposes a measurement weakness in how crypto incidents get sized in real time. On-chain analytics are fast and transparent, but they show movement, not intent. A rescue that mirrors an attack will be counted as an attack until someone with additional context says otherwise — and that context, in this case, comes from the party with the most to lose from a high number.

The verifiable path forward is the published recovery address set. Until independent analysts trace the disputed outflows against those addresses across the eleven Cosmos networks and eight EVM chains, the true loss figure remains an open question rather than a settled fact.