# Attacker Forged a Safe Authorization Check to Drain 114 Ether From an Aave Looping Module

The FlashLoopAdapter asked the caller whether it was allowed, instead of verifying it independently, and two multisig wallets lost about $305,000.

- Published: 2026-10-03T05:57:57.658Z
- Canonical: https://polylog.news/crypto/2026-10-03/attacker-forged-a-safe-authorization-check-to-drain-114-ethe
- Publisher: Polylog (Crypto desk)
- Section: crypto
- Sources: [DeFiHackLabs (proof of concept)](https://github.com/SunWeb3Sec/DeFiHackLabs/commit/2ee728d4e1fe3f860593897a82a9b9595a2cba76), [DeFiHackLabs (second commit)](https://github.com/SunWeb3Sec/DeFiHackLabs/commit/64db8ed91f86650b53bc90eab984d09836876e63), [DeFiHackLabs (SKYDAO proof of concept)](https://github.com/SunWeb3Sec/DeFiHackLabs/commit/2484795d409e7f3e44e00d57d52bba961af0d663)

An attacker drained 114.09 Ether, worth roughly $305,000, from two Safe multisig wallets on Ethereum on October 2 by exploiting a third-party adapter those wallets had enabled as a module. Security researchers at DeFiHackLabs published [a reproducible proof of concept](https://github.com/SunWeb3Sec/DeFiHackLabs/commit/2ee728d4e1fe3f860593897a82a9b9595a2cba76) labelling the root cause broken self-attestation.

The vector was access control, and the flaw was in who answered the authorization question. The adapter's open and close functions did not independently establish that the caller was a genuine Safe. They asked the calling contract whether the adapter was enabled as one of its modules, by calling `isModuleEnabled` on `msg.sender`. The attacker deployed a contract that simply returned true, then directed the adapter to execute arbitrary actions on its behalf. [SlowMist, which traced the transactions](https://www.kucoin.com/news/flash/slowmist-identifies-flashloopadapter-flaw-behind-114-eth-theft-from-two-safe-wallets), reported that the attacker took a Morpho flash loan in wrapped Ether, repaid roughly 1,300 wrapped Ether of debt to unlock the victims' collateral, and kept the surplus. Aave's core lending pools and the Safe wallet contracts themselves were used as designed and were not compromised. No recovery or freeze has been reported.

Two smaller incidents followed the same week. DeFiHackLabs published a proof of concept for [an attack on SKYDAO worth about 59,900 Tether](https://github.com/SunWeb3Sec/DeFiHackLabs/commit/2484795d409e7f3e44e00d57d52bba961af0d663), in which the attacker burned tokens held by the liquidity pair and then forced a reserve resynchronization, moving the pool price before trading against it. A separate case drained roughly 32,000 Tether from a position manager contract that priced vault shares off a live spot pool, allowing an attacker to inflate share value before redeeming.

## What this means

All three incidents share the same root cause: a trust-boundary error, where a contract trusted an input it should have verified independently. The adapter delegated an authorization decision to the party being authorized, and the two smaller contracts delegated a valuation decision to a pool the attacker could move. Audits scoped to a single contract pass in each case, because the contract does exactly what it was written to do. The exposed parties are users of composable modules, meaning anyone who grants an external adapter execution rights over a wallet, and the operators of vaults that read spot prices. Precedent runs from the 2020 bZx oracle manipulations through to the module and keeper compromises of the past two years, and in each wave losses moved toward privileged peripheral code. The likelier path over the coming months is more losses at integration points of this kind, with wallet interfaces beginning to surface module permissions more prominently, unless module registries start requiring independent verification of caller identity.

## What to watch

- Whether Safe or Aave ecosystem teams publish a vetting standard for third-party modules, which would show the integration layer being treated as in-scope for review.
- Any movement of the stolen Ether through mixers or centralized venues, which would indicate whether attribution and freezing are still feasible at this size.
- The share of new proofs of concept in public exploit repositories that cite access control rather than arithmetic or reentrancy bugs.