A malicious operator will control consensus without risking stake (stake-exit lag exploit)

Summary

The non-atomic nature of setSigVerifier and commitValSetHeader will cause a potential loss of security guarantees for networks as a malicious operator can manipulate validator sets after unstaking their funds, avoiding slashing penalties.

Root Cause

In Settlement contract the design choice to separate setSigVerifier and commitValSetHeader functions (latter being public) is a mistake as it allows for a time gap between validator selection and header commitment. This creates a window where an operator can withdraw their stake while still maintaining their voting power.

Specifically, commitValSetHeader in Settlement is marked as public:

https://github.com/sherlock-audit/2025-06-symbiotic-relay/blob/main/middleware-sdk/src/contracts/modules/settlement/Settlement.sol#L292-L324

While setSigVerifier is a separate function:

https://github.com/sherlock-audit/2025-06-symbiotic-relay/blob/main/middleware-sdk/src/contracts/modules/settlement/Settlement.sol#L267-L275

Attack Path

Epoch 1:

1. A malicious operator deposits a large amount of stake to their vault, ensuring it exceeds the quorumThreshold for voting power.
2. The off-chain relay calculates voting powers, creates a sigVerifier using validators belonging to the malicious operator, and calls setSigVerifier to assign these validators for the next epoch.
3. Near the end of Epoch, the operator calls withdraw on their vault to initiate the withdrawal of their stake.

Epoch 2:

1. The operator immediately calls claim on their vault to retrieve all their withdrawn stake, effectively removing their financial exposure.
2. Despite having withdrawn their stake, their voting power is active in the system's state for a short time-span.
3. The operator can craft their own proof using their validators' private keys, which will pass verification since they had enough voting power.
4. When the operator submits this crafted proof to commitValSetHeader, it will be accepted by the system, allowing the operator to control consensus without any stake at risk.

Impact

The network suffers a complete compromise of its security model. The malicious operator can perform any validator action (such as approving invalid transactions or censoring valid ones) without having any stake at risk of being slashed. This fundamentally breaks the economic security assumptions of the protocol, which relies on validators having skin in the game to behave honestly.

Mitigation

Redesign the commitValSetHeader function to be internal (_commitValSetHeader), and create a new public function that handles both setting the signature verifier (optional) and committing the header in a single atomic transaction.

Alternatively apply checkPermission to commitValSetHeader so it can only be executed by the Network's relay service.