Risk desk · 2026-11-06 · Incidents cited only from official post-mortems, issuer statements and regulator filings

Most DeFi losses fall into six categories. Each has a mechanism, a set of checks that can be run before depositing, and at least one well-documented incident where the mechanism failed. This checklist runs through them in the order they tend to matter for a typical position: the code first, then the data the code reads, then the people who can change the code, then the plumbing between chains, then market-driven failures.

Incidents below are cited from the affected project’s own post-mortem or a regulator’s filing. Amounts are given only as those sources state them.

1. Smart contract risk

Mechanism. The contract does something its authors did not intend: a missing check, a rounding error, a reentrancy path, or a bad initialization. Because contracts hold funds directly, a bug is a withdrawal path for anyone who finds it.

Primary-source case. Nomad’s root cause analysis explains that a contract upgrade on 2022-06-21 changed how the bridge’s Replica contract checked messages. A Replica initialized with a zero root marked that root as valid, and since unproven messages also map to a zero root, the check that should have rejected them always passed on those Replicas. In plain terms, any message could be processed without proof. Nomad’s later recovery post states the bridge “was hacked for more than $186M” on 2022-08-01 and that over 300 unique addresses participated in the exploit.

Checks.

  • Audits exist, are recent, and cover the deployed commit (compare the commit hash in the audit to the verified source).
  • Upgrades after the last audit are themselves reviewed. Nomad’s failure came from an upgrade.
  • Contract source is verified on the block explorer.
  • A bug bounty exists with a payout proportional to funds at risk.
  • Time in production with meaningful value at stake (code that has held funds for years has been probed more).
  • The protocol depends on as few external contracts as possible, and those are also audited.

2. Oracle risk

Mechanism. Lending markets, perps and many vaults need a price. If that price can be moved, an attacker can make collateral look more valuable than it is, borrow against it and leave the protocol with bad debt. ethereum.org’s oracle page puts it simply: “Data from an oracle must be correct for a smart contract to execute correctly.”

Primary-source case. The CFTC’s January 2023 complaint against Avraham Eisenberg alleged that he took large positions in a MNGO/USDC swap on Mango Markets, then bought MNGO on three exchanges that fed the platform’s oracle. The CFTC alleged the oracle price rose more than 13-fold in 30 minutes and that he used the inflated position value as collateral to withdraw over $110 million. These are the regulator’s allegations as filed.

Checks.

  • Identify every price source the protocol reads, per asset.
  • For each, check how much liquidity sits behind it. A token with thin spot markets is an oracle risk wherever it is accepted as collateral.
  • Check whether the protocol uses a time-weighted average, a median of several sources, or a single spot read.
  • Look for caps: borrow caps or supply caps on illiquid collateral limit how much a manipulated price can extract.
  • Check what happens if the oracle stops updating (stale-price handling).

3. Governance and admin-key risk

Mechanism. Someone can usually change the contracts: a multisig, a token-holder vote, or both. That power can fix bugs, and it can also redirect funds if the keys are compromised, the signers collude, or the voting process can be captured.

Primary-source case. Beanstalk’s post-mortem (2022-04-19) says that on 2022-04-17 an attacker “used a flash loan to exploit the protocol’s governance mechanism,” and that approximately $77 million was stolen from its liquidity pools. Borrowed capital, held for a single transaction, was enough to pass a malicious governance action.

Checks.

  • Who holds upgrade and pause rights? A named multisig, a DAO, or a single key?
  • For a multisig: how many signers, what threshold, and are signers independent and public?
  • Is there a timelock between approval and execution, and is it long enough for users to exit?
  • Can voting power be borrowed or flash-loaned (snapshot timing, vote delays)?
  • What can admins do without a vote (pause, change oracles, change fee recipients, mint)?
  • For rollups, ETHBEAT (formerly L2BEAT) publishes stage requirements, including the exit window given to users before upgrades. The same logic applies to any protocol.

4. Bridge risk

Mechanism. A token bridged from another chain is an IOU against assets locked in a contract on the origin chain. If the bridge’s verification fails, an attacker can withdraw the locked assets or mint unbacked IOUs. Bridges concentrate value from many users into one set of contracts and one set of verifiers.

Primary-source case. Sky Mavis stated that on 2022-03-23 its Ronin validator nodes and Axie DAO validator nodes “were compromised resulting in 173,600 Ethereum and 25.5 million USDC drained from the Ronin bridge,” that the attack was socially engineered, and that the root cause was “the small validator set.” The bridge is covered in more depth, with Nomad and Wormhole, in cross-chain bridges.

Checks.

  • How are cross-chain messages verified: a multisig or validator committee, light clients, or validity proofs?
  • How many verifiers, at what threshold, and how independent are they?
  • Is the bridged asset the issuer’s canonical version on that chain or a third-party wrapper?
  • Are there rate limits on withdrawals?

5. Liquidation risk

Mechanism. Borrowing against volatile collateral means a price drop can trigger liquidation, where a third party repays part of the loan and takes collateral at a discount. In a fast market, liquidations cluster, gas prices spike and collateral may be sold well below the threshold price. Liquidation is a designed feature, not a bug, but it converts temporary price moves into permanent losses for the borrower.

Hypothetical illustration: a borrower posts collateral worth $10,000 and borrows $6,000 in a market with an 80% liquidation threshold. The position becomes liquidatable when collateral value falls to $7,500, a 25% drop. If the collateral is itself a liquid staking token that trades at a discount during stress, that discount counts toward the 25%.

Checks.

  • Liquidation threshold and penalty for each collateral type, from the protocol’s docs.
  • Which oracle sets the liquidation price, and how fast it updates.
  • How correlated the collateral is with the borrowed asset.
  • Whether the protocol has a backstop (insurance fund, reserve) for bad debt and who absorbs losses if it is exhausted.

6. Depeg risk

Mechanism. Stablecoins, liquid staking tokens and wrapped assets are all designed to track something. If confidence in redemption breaks, the token trades below its reference, and every protocol that priced it at par inherits the gap.

Primary-source case. TerraUSD, marketed as an algorithmic stablecoin, lost its peg in May 2022. The SEC’s press release 2024-73 announced Terraform Labs and Do Kwon would pay more than $4.5 billion following a fraud verdict, and says the depeg erased about $40 billion in market value “almost overnight.”

Checks.

  • What backs the token, and who can redeem it directly (often only whitelisted institutions)?
  • For fiat-backed tokens, what reserve reports exist and how often. In the US, the GENIUS Act requires payment stablecoin reserves of at least 1:1 once effective (the earlier of 2027-01-18 or 120 days after final regulations).
  • For LSTs, how long the protocol’s withdrawal queue takes, since queue length sets how far the market price can drift.
  • Which pools and lending markets hold the token, since depeg contagion runs through them.

How the categories interact

The six categories rarely fail alone. The Mango case combined an oracle weakness with a lending feature that let inflated collateral be borrowed against. Beanstalk combined governance design with flash-loan liquidity from elsewhere in DeFi. A depeg in a liquid staking token becomes a liquidation event in every lending market that accepts it. When scoring a protocol, it helps to trace one failure through every dependency it touches, rather than rating each category on its own. A protocol with excellent code but a single-signer admin key, or a strong audit history but collateral priced from one thin pool, carries the risk of its weakest dependency.

Cross-cutting checks

  • Front end. The website is separate software. Domain hijacks and injected scripts can request malicious signatures while the contracts are untouched.
  • Approvals. Unlimited token approvals to a contract persist until revoked.
  • Concentration. If a few wallets hold most of the deposits, one exit can change pool depth, utilization and rates for everyone else.
  • Jurisdiction. Many derivatives front ends restrict US users in their own terms, and the CFTC has charged DeFi operators for offering leveraged retail commodity transactions to US persons.
  • Insurance. FDIC insurance does not apply to tokens or smart-contract deposits.

A checklist reduces surprises; it does not make any protocol safe, and working through one is not an endorsement of the result. The longer due-diligence process, including token unlocks and team checks, is in how to research a crypto project. The mechanics behind several of these risks are in what is DeFi and yield farming.

Sources