oak_techniques: [OAK-T9.003] spec_id: oak-detection-T9.003 version: 0.1.0 maturity: stable maintainer: "@iZonex" license: Apache-2.0 scope: | Detect exploitation of governance-contract code-level flaws — the contract itself converts a vote into protocol state incorrectly. Covers three sub-classes: same-block flash-borrow vote against a no-snapshot/no-timelock governance (Beanstalk); malicious-proposal payload with hidden delegatecall / state-replacement (Tornado Cash May-2023); spot vote-accumulation by a delegate consortium without governance-process resistance (Compound Proposal-289). Excludes: T16.001-005 (voting-power abuse against a contract that operates as designed); T9.002 (flash-loan capital — frequently chained, separate primitive); legitimate snapshot-and-timelock governance flow (suppressed via canonical pattern checks). data_sources: [tx_call_trace, governance_events, contract_bytecode, proposal_calldata, token_transfer_events, funder_graph] detection_logic: description: | Four orthogonal paths. PATH A (flash-loan vote): same-tx flash-loan funding + governance-token acquisition + vote/execute on a non-snapshot governor. PATH B (proposal-payload static): proposal calldata performs `delegatecall` to non-canonical target, sets privileged roles, or selfdestruct/upgrade-to-attacker-impl. PATH C (governance design): governor contract reads spot voting-power and has emergency-execution path bypassing the lifecycle. PATH D (spot-accumulation): single voter's weight on a treasury-affecting proposal exceeds historical-participation envelope by ≥ k×. pseudocode: | PRIVILEGED_PROPOSAL_TARGETS = {treasury, mint_role, upgrade, role_admin, owner_set, fee_router} # PATH A — flash-loan vote acquisition (Beanstalk-class) for each tx T touching a governance contract G: flashloan ← exists frame f in T : selector(f) ∈ FLASHLOAN_PROVIDERS acquire ← Σ T.token_in[gov_token(G)] > flashloan_acq_ratio × circulating(gov_token(G)) vote_or_exec ← any(f.selector ∈ {castVote, queue, execute, emergencyCommit} and f.contract == G for f in T.frames) if flashloan and acquire and vote_or_exec and not snapshot_used(G, T): emit(PATH_A, tx=T.hash, governance=G, borrowed_usd=flashloan_borrow_usd(T), severity=critical) # PATH B — malicious proposal payload (Tornado-2023-class) for each proposal P submitted to G: decoded ← decode_calldata(P.payload, abi=G) for op in decoded.ops: target_unknown ← op.target ∉ canonical_targets(G) risky ← op.kind in {delegatecall, selfdestruct, upgradeTo} or touches(op, PRIVILEGED_PROPOSAL_TARGETS) if target_unknown and risky: emit(PATH_B, governance=G, proposal=P.id, op=op, severity=critical) # PATH C — governance-design surface (pre-deployment / one-shot) if reads_spot_voting_power(G) and not has_snapshot_checkpoint(G): emit(PATH_C, governance=G, surface="no-snapshot", severity=high) if has_emergency_execution_path(G) and not gated_by(emergency_path(G), timelock_allowlist): emit(PATH_C, governance=G, surface="emergency-bypass", severity=high) # PATH D — spot-accumulation envelope (Compound-289-class) for each proposal P touching PRIVILEGED_PROPOSAL_TARGETS: voted_weight ← Σ vote.weight for vote in P.votes top1 ← max(vote.weight for vote in P.votes) median_hist ← median(top_voter_weight(p) for p in recent_proposals(G)) if top1 / voted_weight > top_voter_share and top1 > spot_accum_multiplier × median_hist: cluster ← funder_graph_cluster({v.voter for v in top_k(P.votes, k=10)}, hops = trace_hops) if shared_funding(cluster): emit(PATH_D, governance=G, proposal=P.id, top1_share=top1/voted_weight, cluster, severity=high) parameters: flashloan_acq_ratio: { type: number, default: 0.10 } # ≥10% supply via flash loan canonical_targets: { type: object, default: {} } # per-governor allowlist timelock_allowlist: { type: list, default: [] } top_voter_share: { type: number, default: 0.30 } # single voter > 30% of voted weight spot_accum_multiplier: { type: number, default: 5 } # vs historical median top voter trace_hops: { type: integer, default: 3 } output_alert: [oak_technique, detection_path, severity, chain, governance_address, proposal_id, tx, evidence] test_fixtures: positive: - 2022-04-beanstalk # PATH A — flash-loan + emergencyCommit - 2023-05-tornado-cash-governance # PATH B — hidden selfdestruct/state-replacement - 2024-07-compound-vote-takeover # PATH D — Proposal 289 spot-accumulation - 2022-07-audius # PATH B-adjacent — storage-collision via init - 2024-03-curio # PATH A/C — small-protocol flash-vote negative: - "Compound Governor Bravo proposal — checkpoint snapshot + 2-day timelock + canonical execution targets, must not flag" - "Snapshot-based off-chain vote that signs but does not execute on-chain" false_positive_modes: - legitimate proposals queued/executed by canonical multisig + timelock — allowlist via timelock_allowlist + canonical_targets - proposals that delegatecall a known canonical helper (executor library) — allowlist by codehash in canonical_targets - governance-token large transfers reflecting treasury operations or vesting unlocks (not vote acquisition) — PATH A's vote_or_exec gate filters - high-participation proposals where top voter is a known long-term delegate above the threshold — historical-envelope check should reflect this; calibrate spot_accum_multiplier per governor mitigations: [OAK-M02, OAK-M05, OAK-M11, OAK-M16, OAK-M17, OAK-M32, OAK-M33, OAK-M34, OAK-M35, OAK-M39] reference_implementations: - { target: forta-bot, chain: evm, url: "" } - { target: oz-defender-sentinel, chain: evm, url: "" } - { target: tally-analytics, chain: evm, url: "" } - { target: dune, chain: evm, url: "" } - { target: tenderly-virtualnet, chain: evm, url: "" }