DeFi Governance · Research Brief
April 2026

The governance handbrake.

How security councils solved DeFi's speed-versus-decentralization problem — and the new vulnerabilities they introduced.
01 / 10
Anay, Atharva, Ved
02 Part I · The Baseline Problem Compound · Sept 2021

A pure DAO cannot outrun its own bug.

How $80M drained
  1. Proposal 62 shipped a new COMP distribution model — approved by DAO vote.
  2. A single flawed operator (> where >= was needed) let the contract release COMP without checking supply caps.
  3. The moment the timelock ended and the upgrade went live, the contract began paying out 280,000 COMP per transaction to anyone who claimed.
  4. Developers spotted it in minutes. But the only fix was another proposal — which triggered another 7-day timelock.
  5. For a week, the community watched the drain live on-chain. $80M+ in COMP was mistakenly claimed before the patch landed.
The drain
$80M
vs.
The fix
7 days
speed of governance
Decentralized governance speed is incompatible with the speed of a smart-contract exploit.
03 Part II · The Solution The Security Council

A small, elected committee with super-admin keys — used rarely.

A multi-sig of trusted signers who bypass DAO voting delays in an emergency. Peacetime: routine maintenance. Wartime: instant intervention.

Mechanism
Gnosis Safe multi-sig
Bypass
Emergency role on the TimelockController — delay = 0
Authority
Pause · upgrade · execute arbitrary transactions
Signature threshold
9 of 12
Nine distinct private keys must sign. No single member, no small subset, can act alone.
04 Part II · Mechanism Timelock Bypass

Two routes to a state change.

Standard DAO route
Democratic.
Deliberate.
01Submit proposalonchain
02Community vote3 days
03Mandatory timelock2–4 days
04Execute state change—
Time to act
7 days
Security Council route
Emergency.
Instant.
01Detect anomalymonitoring
02Coordinate signersminutes
03Gather 9 signatures—
04Execute with delay = 0one block
Time to act
< 1 hour
05 Part II · Available Powers Circuit Breakers

Three emergency powers, ordered by blast radius.

I
Circuit breaker
Protocol pause.
Instantly freeze deposits, withdrawals, or trading on specific contracts. Least destructive — stops the bleed without moving funds.
II
Hot patch
Contract upgrade.
Force an immediate logic change on a proxy contract to patch a vulnerability. Rewrites behaviour in a single block.
III
Nuclear option
Arbitrary transaction.
Execute custom logic: intercept funds, blacklist malicious addresses, revert specific actions. Full sovereign reach over the protocol.
06 Part III · In Practice Kelp DAO · April 2026

Where Compound failed, Arbitrum's council acted.

How $70M was saved
  1. An exploit in Kelp DAO's restaking contracts began draining user deposits — ~$300M siphoned across DeFi positions.
  2. The attacker bridged 30,766 ETH (~$70M) to Arbitrum as a holding wallet before attempting to launder it.
  3. Arbitrum's Security Council detected the incoming funds. They skipped the 7-day vote and mobilised in minutes.
  4. With 9 signatures, they executed a single atomic transaction on Ethereum mainnet that intercepted the ETH mid-flight — before the attacker could move it again.
  5. Funds were routed to a DAO-controlled recovery address. Zero network downtime, zero collateral damage.
Case study · Exploit & recovery
Kelp DAO
$70M recovered
Exploit total
~$300M across positions
ETH intercepted
30,766 ETH on Arbitrum
DAO vote wait
0 days — bypassed
Network downtime
None
Mechanism
Atomic bridge upgrade on L1
07 Part III · Why this matters Inside the rescue transaction

The exact four moves that recovered $70M.

Kelp's recovery wasn't magic — it was the council's powers used surgically. This slide shows how upgrade · impersonate · transfer · revert compose into a single transaction that no pure-DAO governance model could have authored in time.

Step 01
Upgrade.
inbox.upgradeTo(temp)
Temporarily replace the Ethereum→Arbitrum bridge implementation with a custom one.
Step 02
Impersonate.
sendL2Msg(hacker)
Inject a function that forges cross-chain messages as if sent by the hacker's address.
Step 03
Transfer.
eth.transfer(30_766)
Move 30,766 ETH from the hacker's Arbitrum wallet to a DAO-controlled recovery address.
Step 04
Revert.
inbox.upgradeTo(orig)
Restore the bridge to its original implementation. Backdoor closed in the same transaction.
All four steps, one atomic transaction — no window for collateral damage.
$70M+ saved · zero downtime
08 Part IV · The New Frontier Open vulnerabilities

The council model has its own attack surface.

01 · Economic
Flash loan governance attack.
An attacker borrows massive voting power in one block, passes a malicious proposal, and repays the loan. If council reaction lags or timelocks don't guard flash votes, the DAO is legally hijacked in seconds.
Precedent — Beanstalk Farms · $76M
02 · Operational
Key compromise honeypot.
Concentrating control in 12 signers creates a single high-value target. A sophisticated actor — state-sponsored or otherwise — who compromises signer devices inherits the protocol.
03 · Legal
Subpoena & coercion.
Council members are known individuals with the power to freeze or move assets. That makes them uniquely vulnerable to subpoenas, sanctions, and direct regulatory pressure.
04 · Philosophical
Decentralization theater.
If 9 people can impersonate any user and seize funds — even for good reasons — is the network actually decentralized? The trust model quietly shifts from trust the code back to trust these humans.
09 Part IV · The Tradeoff No Free Lunch

Every council is a bet on which failure mode you'd rather survive.

Side A · pure decentralization

Censorship resistance.

  • No humans to coerce, subpoena, or compromise.
  • Fully trustless — the code is the only authority.
  • Fatal to speed: a 7-day timelock is a 7-day drain.
Side B · council model

Rapid response.

  • Instant emergency intervention — minutes, not days.
  • Proven recovery of funds mid-exploit.
  • Trust shifts back to identifiable humans with keys.
●  Conclusion
10 / 10

Trust the code, or trust the keys.

Part I
Pure decentralization cannot outrun its own bugs.
Part II
Councils are the industry's pragmatic answer — a handbrake with god-mode keys.
Part III
Kelp DAO proved the model works: $70M+ recovered, zero downtime.
Part IV
The handbrake is also a honeypot. The trust model has moved — not disappeared.
Discussion welcome
Thank you