“Audited” is the most overworked word in DeFi. It appears on landing pages next to a logo, in pitch decks, and in the replies of anyone accused of shipping something dangerous. What it actually means is narrower than most users assume, and understanding the gap is one of the cheapest pieces of security you can buy.
What an audit is
A smart contract audit is a time-boxed review of source code by a team that looks for ways to break it. The output is a report: a scope statement, a threat model, a list of findings with severities, and usually a note on which findings were fixed.
Three properties define it:
- Bounded by scope. The auditors review what they were paid to review. If the proxy, the keeper bot, the frontend, or the deployment script was out of scope, it was not audited, whatever the badge says.
- Bounded by time. Findings apply to a commit. Any line changed afterwards is unreviewed until the next engagement.
- Bounded by imagination. Auditors test against known failure modes and their own experience. Novel economic designs can be internally consistent and still exploitable.
The glossary entry on smart contract audits covers the vocabulary; this piece is about how to read the result.
What a smart contract audit reliably checks
Across firms, the same classes dominate findings:
Access control. Who can mint, pause, upgrade, withdraw, change fees. The single most common catastrophic bug is a privileged function with no guard, or a guard that checks the wrong role.
Arithmetic and decimals. Rounding in the protocol’s favour or the user’s, decimal mismatches between tokens, and the accumulation of small errors into a drainable imbalance.
Reentrancy and call ordering. External calls that hand control to attacker code before state is settled. Modern Solidity guard rails help, but cross-contract and cross-function paths still produce findings every year.
Oracle and price assumptions. Whether a price can be moved within one block, whether a stale feed is rejected, whether a TWAP window is long enough. An audited contract with a manipulable oracle is audited and exploitable, which is why oracle manipulation keeps recurring.
Upgrade and initialisation. Whether an attacker can call initialize, whether the implementation behind a proxy can be swapped by the wrong address, whether storage layouts collide. The upgradeable contract pattern is where a lot of quiet risk lives.
Token standard compliance. Fee-on-transfer tokens, rebasing balances, non-standard return values: contracts that assume clean ERC-20 behaviour break in predictable, exploitable ways.
What it cannot do
The four gaps that matter, in order of how often they explain a post-audit loss:
1. Keys are not code. An audit says nothing about how the deployer’s private key is stored, whether the upgrade multisig has four signers or one, or whether the keeper bot’s API key is in a public repository. The Lighter incident in November 2026 is the clean illustration: the contracts worked exactly as designed, and $10.4 million left through a leaked signing key.
2. The code can change. Upgradeable contracts run different bytecode after a governance vote. An audit of version 1 tells you about version 1. If nobody re-audits the diff, the report becomes a historical document attached to a live system.
3. Economic attacks are not bugs. Flash-loan-driven governance capture, incentive misalignment, liquidation cascades: these can be “correct” code and still extract value. Auditors flag them when they see them; the design is not required to be safe.
4. Runtime is out of scope. An audit does not monitor. It does not notice a compromised admin key at 3am, an anomaly in outflow, or a bridge accepting a forged receipt because the signature verification logic trusted a field it should have hashed. Our bridge security explainer has several such cases.
How to read a report in five minutes
Open the report and look for these, in this order:
- The commit hash. Compare it with the deployed contract’s verified source. If they differ and there is no follow-up audit, the “audit” covers code that is not running.
- The date relative to the last upgrade. An audit from 18 months ago attached to a protocol that has upgraded four times is marketing.
- Scope. Which contracts, which files, which out-of-scope pieces. Watch for “frontend and infrastructure excluded”, because that is where key leaks happen.
- High and critical findings, and their status. “Resolved” with a referenced fix commit is good. “Acknowledged” means the team decided to ship anyway; read the reasoning.
- Informational findings. These are usually style and gas notes, but they occasionally contain a design assumption the auditors found uncomfortable. Skim them.
If the site shows a badge but no report, treat it as unaudited. If it shows a report but no commit hash, treat it as unaudited with better typography.
Red flags in how audits are marketed
- “Audited by” with no link. The report is the artefact; the logo is not.
- Auditing an older version. Common after a rushed upgrade, and the most dangerous pattern in the list.
- “Zero critical findings” used as a safety claim. Zero critical findings means the auditors’ process did not surface one, not that none exist.
- Audit of a fork’s upstream. A fork inherits the code, not the assurance; the changes are the risky part.
- No mention of key management. If the deployer key is a hot wallet, the audit’s scope is the least of your problems.
What to do as a user
You cannot audit the code, but you can check the four facts above in a couple of minutes, and you can size your exposure accordingly: new protocol, unaudited diff, or a single-signer admin key means small position, separate wallet, and an approval you will revoke later.
And keep the framing straight. An audit buys you a lower probability of one class of failure. It does not buy you custody, uptime, honest governance, or a refund. Those are separate products with separate prices, and most of them are not for sale.
Limitations
This is a user’s guide, not an auditor’s. It does not cover formal verification, fuzzing coverage, or how to evaluate a specific firm’s methodology, and firm quality varies widely.
It also assumes the report is honest. Occasionally it is not: reports have been commissioned, softened, or published with findings quietly removed. When a protocol’s only evidence of diligence is a logo and a PDF with no commit hash, the correct response is not to read more carefully. It is to assume nothing.
Frequently Asked Questions
Does a smart contract audit guarantee the contract is safe?
No. An audit is a point-in-time review of a specific commit against a stated threat model. It reduces the probability of known bug classes; it cannot prove absence, and it says nothing about the keys that can upgrade or pause the contract. Treat it as evidence of diligence, never as a guarantee.
What does an audit actually look at?
Typically: access control on privileged functions, arithmetic and decimals handling, reentrancy and external call ordering, oracle and price-feed assumptions, upgrade mechanics and initialisation, token standard compliance, and the economic invariants the design depends on. Most reports also include automated tooling output plus manual review notes.
Why do audited protocols still get hacked?
Four reasons recur: the bug was outside the scope, the code changed after the audit, the exploit was economic rather than a code bug, or the loss came from an operational failure such as a leaked key. Our incident files contain examples of all four, including a bridge whose flaw was in its proof checking rather than its Solidity.
How can I tell whether an audit is worth anything as a user?
Check four things: does the report exist and is it public, does the commit hash in the report match the deployed contract, how old is it relative to the last upgrade, and are the high-severity findings marked resolved with a follow-up commit. If any of those is missing, the audit is decoration.
How much does a smart contract audit cost?
For a small DeFi contract, typical engagements run from a few thousand to tens of thousands of dollars depending on scope and firm reputation; large protocols pay more for continuous review. Cost is not a quality signal by itself: scope, auditor experience and whether the report is public tell you more.