iktok90-design/ai-smart-contract-auditor/centralization-risk

Centralization risk audit

Why this matters

Centralization findings are non-exploits in the traditional sense, but they're the most common reason audits flag a protocol as "rugpullable". The protocol can be the attacker. Investors deserve a clear inventory.

When this applies

• Every privileged function (mint, burn, set-fee, set-oracle, upgrade, pause, recover, sweep, blacklist)
• Multi-sig configuration
• Timelock configuration
• Upgrade authority (UUPS _authorizeUpgrade)
• Emergency-withdraw / emergency-pause
• Token mint caps + upgradability of caps

Detection patterns

Single-key admin (HIGH)

contract X is Ownable {
    function upgradeTo(address impl) external onlyOwner { ... }
    function setFee(uint256 fee) external onlyOwner { ... }
    // ← one key, can pull rug
}

Findings:

• "Owner can upgrade implementation immediately, with no timelock or multi-sig."
• "Single key controls treasury."

No timelock (HIGH)

onlyOwner upgrades / fee changes / oracle changes that take effect immediately. Add a Timelock (OZ TimelockController), document the delay.

Pause-without-time-cap (HIGH)

pause() with no auto-unpause. Owner can permanently freeze user funds.

Mint without cap (HIGH)

function mint(address to, uint256 amt) external onlyMinter { _mint(to, amt); }   // ← no cap, inflation risk

Blacklist without on-chain criteria (HIGH)

A blacklist(address) function with no public criteria = "the protocol's discretion to freeze you".

Oracle controlled by single admin (HIGH)

function setOracle(address o) external onlyOwner { oracle = IOracle(o); }

Admin → switches to malicious oracle → liquidates everyone.

Sweep-any-token without allowlist (HIGH — also in [[access-control]])

recoverERC20(token) that can sweep user deposits.

Upgrade authority over user funds, no exit window (HIGH)

Users should have a window after upgrade announcement to withdraw if they disagree. Without it, governance can rug by upgrade.

"Renounced ownership" theater (HIGH)

Ownable renounced but DEFAULT_ADMIN_ROLE retained. Or admin role transferred to a smart contract that has its own backdoor.

EOA multisig threshold of 1 (HIGH)

Marketed as "multi-sig" but threshold == 1 defeats the purpose.

Validator / sequencer centralization (HIGH for bridges, L2s)

Single sequencer / centralized validator set. Document for bridges + L2 deployments.

Treasury without time-locked spend (HIGH)

Owner can drain treasury immediately.

selfdestruct reachable by admin (MEDIUM-HIGH)

See [[selfdestruct-eip6780]].

Severity rubric

Output format

Centralization findings should produce a "Trust report" section in /report:

Admin powers:

- setOracle — can replace price oracle → drain liquidations (no timelock).

- upgradeTo — can swap implementation (UUPS, no timelock).

- recoverERC20 — can sweep any token, including user deposits.

- pause — can freeze withdrawals indefinitely (no auto-unpause).

- setFee — can set fee to 100% (no cap).

Mitigations in place:

- Owner is a Gnosis Safe at 0x… with 3/5 threshold.

- 24h Timelock at 0x… mediates setOracle and upgradeTo.

Trust assumption: Users must trust the multi-sig signers and the 24h timelock window.

Remediation patterns

• Multi-sig (3+ of N) for admin role.
• Timelock (≥48h for upgrades, ≥24h for params).
• Per-function caps on what admin can change (fee ≤ 5%, mint ≤ 10% of supply).
• Auto-unpause after N hours.
• Withdraw-only mode on pause.
• Public "trust assumptions" doc — every admin function listed.

False-positive notes

• A pre-launch protocol with single admin pending timelock setup is normal — flag with recommendation, not critical.
• Test networks / staging contracts are fine to centralize.

Related

• [[access-control]]
• [[initialization]]
• [[delegatecall-risks]]