A DeFi protocol is a set of smart contracts that holds deposits and moves them according to fixed rules without a bank or broker in between, and the sensible first act for a newcomer is to read one before depositing in it. DefiLlama tracks more than 7,000 of them, the sector held $95.42 billion on September 26, 2026 by Inside Deep Tech's count, and in the third quarter of 2026 $1.2 billion was lost across 247 exploits. This page takes a reader from the definition to a written one-paragraph account of a single protocol: what it does, where its yield comes from, what has gone wrong with it and who can change its rules.

Code that holds the money

Everything else on the subject rests on one idea. A DeFi protocol is a program that has custody of other people's assets and treats them exactly as its code says. The program is a smart contract: software deployed on a blockchain that executes as written when someone sends it a transaction. The U.S. Congressional Research Service, in its primer on decentralized finance (issued March 16, 2026, updated April 17, 2026), defines DeFi as financial services conducted through such protocols without centralized intermediaries. It adds a stricter test: to count as decentralized, the protocol must be autonomous, permissionless and transparent, with no central party in control. Many protocols pass part of that test and not all of it.

A lending protocol shows the idea at work. A depositor puts stablecoins into a shared pool. A borrower who wants to draw from that pool must first lock up collateral (assets pledged against the loan) worth more than the loan itself. The practice is called overcollateralization, and it is how the protocol lends to strangers without checking anyone's credit. The interest rate is not set by a loan officer. It moves automatically with utilization, the share of the pool currently lent out: the more is borrowed, the higher the rate. If the collateral's value falls below a set threshold, the contract sells it to repay the loan. That sale is a liquidation, and nobody has to approve it. The depositor's yield is simply the borrowers' interest.

The largest protocols are usually ranked by TVL, total value locked: the dollar value of assets sitting in a protocol's contracts. On DefiLlama's late-September 2026 snapshot, Lido led at roughly $26.46 billion, followed by Aave at about $19.145 billion. Trackers count slightly differently, so the same protocol can carry somewhat different figures on different sites. A size figure is not a safety rating either. As one research page on the rankings puts it, "TVL measures size and liquidity, not safety or yield."

The phrase "just code" also needs a qualification. Plenty of protocols keep admin keys, privileged credentials that can change or pause the contracts. As later sections show, those keys have become the favored point of attack. Who can change the code is also the question of governance, which on many protocols means a vote by holders of the protocol's token. A separate page here sets out how the code differs from the organization that votes on it.

The protocols fall into a handful of types. Each holds money for a different reason and pays for a different reason. The figures below are as the sources gave them, on the dates shown.

Type What the code does Where any yield comes from Size, as reported Risk the sources stress
Decentralized exchange (DEX) Swaps one token for another against a liquidity pool, priced by a formula (an automated market maker, usually x × y = k) rather than an order book A share of swap fees, paid to those who supply the pool 975 protocols, $12.20B combined (DefiLlama, late Sept 2026) Impermanent loss for liquidity providers
Lending Pools deposits and lends them against overcollateralized loans Interest paid by borrowers 587 protocols; Aave about $19.145B, Morpho $11.131B (DefiLlama, late Sept 2026) Liquidation; price-feed manipulation
Liquid staking Stakes assets such as ETH and issues a tradable receipt token in return Staking rewards on the underlying asset 281 protocols; Lido about $26.46B (DefiLlama, late Sept 2026) Dependence on the receipt token keeping its backing
Restaking Pledges already-staked ETH to secure additional services beyond the base chain Rewards for securing those additional services EigenCloud $6.982B (Inside Deep Tech, Sept 21 2026) One deposit carrying several layers of dependency
Stablecoins Issues tokens meant to hold a fixed dollar value, backed by fiat reserves, by crypto collateral or (rarely) by an algorithm Not a yield product in itself Crypto-backed: 100 protocols, $8.37B (DefiLlama); USDT $183.1B supply (Idealogic, Aug 2026) Loss of the peg if backing fails
Bridges Move assets between blockchains, often by locking them on one chain and issuing a wrapped version on another None by design; a bridge is transport 47 protocols, $8.85B (DefiLlama, late Sept 2026) 61 exploits and $3.3B lost, about 42% of DeFi theft (Idealogic)
Derivatives Perpetual DEXs such as dYdX offer leveraged trading; options protocols are tracked separately Not set out in the sources used here Hyperliquid $6.3B (Idealogic, Aug 2026) Leverage amplifies losses

Two features of the table matter more than any single row. The first is concentration. In one count of the top 100 protocols, the ten largest held 48.4% of the total, so a newcomer will meet a small number of names repeatedly. The second is that the categories lean on one another. A lending market accepts liquid staking tokens as collateral, and a bridge supplies the wrapped asset a DEX trades. A failure in one category therefore rarely stays inside it.

What the exercise requires

Very little. The first result described below needs a web browser, three public sources and somewhere to write a paragraph. The sources are DefiLlama's protocol rankings for size and category, Rekt for the record of exploits above roughly $3 million, and the chosen protocol's own documentation.

It needs no wallet, no deposit and no transaction fee, because nothing is signed and nothing is sent. That is deliberate. Every mistake on the list further down costs money only once a wallet is connected, and the reading is what tells a newcomer whether connecting one is worth it. A single sitting is enough for one protocol.

Reading one protocol before depositing in it

The first real result is a short written account of one protocol, built from its own figures and history. The steps below use lending as the example because its mechanism is the easiest to check. The same steps work for any category.

Building a one-paragraph account of a single protocol

  1. Open the rankings and find the category

    Go to DefiLlama's protocols page and filter by category, choosing Lending. What it should show: a ranked list of lending protocols with their TVL. On the late-September 2026 snapshot, Aave led at about $19.145 billion and Morpho followed at $11.131 billion.

  2. Pick one protocol and write down four facts

    Note its category, the chains it runs on, its TVL and the date of that figure. What it should show: a protocol page with a TVL history. If the figure climbed sharply, check whether prices did too. Sector TVL rose 37.9% between June 30 and September 26, 2026, while ETH rose roughly 67%, so much of that growth was repricing rather than new deposits.

  3. State where the yield comes from, in one sentence

    Use the protocol's documentation to name the source. For a lending market the answer is borrowers' interest, set by utilization. What it should show: a sentence that names who pays. If the documentation describes the yield only as a percentage and never says who pays it, write that down; it is a finding in itself.

  4. Search the incident record

    Search Rekt for the protocol's name and for any bridge, price feed or collateral token it depends on. What it should show: either no listed incident, or a post-mortem naming a cause. The causes recur: price-feed manipulation (Nostra on Starknet, $3.5 million, Sept 23, 2026; Tectonic on Cronos, $120.4 million, Sept 14, 2026), bridge logic flaws, and governance votes passed on thin turnout.

  5. Find out who can change the rules

    Look in the documentation for a governance process, an admin key or a multisig (a wallet that needs several signers to act). What it should show: a named mechanism, such as a token vote with a forum stage, or a team-held key. Term Labs lost $8.5 million on Aug 26, 2026 to a governance proposal that passed with minimal voter participation, so turnout belongs in the note as well.

  6. Write the paragraph

    Combine the four facts, the yield source, the incident record and the governance mechanism. What it should show: five or six sentences that a stranger could check against the links. That paragraph is the first result.

The fifth step is the one most newcomers skip, and it is the one that most often decides the outcome. The way token votes and forum proposals approve changes at the large lending protocols is laid out on the page about who approves contract changes.

Where newcomers go wrong

Treating size as safety. A large TVL figure says many people have deposited. It does not say the deposits are safe. The sign of this mistake is a decision that cites only rank. The remedy is the incident search in step four.

The strongest objection deserves a full hearing. The largest protocols have run longest, attracted the most scrutiny and survived the most stress, so size is not a useless signal. That is fair, and it is why the steps begin with the rankings. But size does not isolate a protocol from its neighbors. When the KelpDAO bridge was breached on April 18, 2026 and 116,500 unbacked rsETH were minted, Aave's TVL fell $6.28 billion within 48 hours and nine protocols froze markets. Aave's own code was not the cause. A large protocol tends to inherit the risks of everything it accepts as collateral.

Reading an audit as a guarantee. An audit is a review of code at a point in time. Drift Protocol lost $285 million on April 1, 2026 in 128 seconds through a compromised admin key. A 2024 audit had flagged the mechanism involved but rated it only "informational." KelpDAO had also been audited. The sign of this mistake is the word "audited" in a decision with nothing after it. The remedy is to read the audit's list of findings and check what was left unfixed.

A protocol can pass a flawless code audit and still lose millions because of a compromised admin key.

Ronghui Gu
Co-founder, CertiK

Supplying liquidity to a DEX without pricing impermanent loss. Impermanent loss is the gap between the value of assets deposited in an AMM pool and the value of the same assets simply held. It appears when the two tokens' prices diverge, and it becomes permanent if the provider withdraws while they are still apart. Swap fees earned are kept either way. The sign of this mistake is a withdrawal worth less than the original holdings would be. Chainlink's explainer gives the scale.

How impermanent loss grows with a price move

Using Chainlink's example: a provider deposits 1 ETH and 1,000 USDC into a constant-product pool, so ETH is priced at 1,000 USDC. The table shows the loss against simply holding, for each change in the price ratio between the two assets.

Price ratio moves 1.25x0.6% loss vs holding
Price ratio moves 1.5x2.0% loss vs holding
Price ratio moves 2x (ETH doubles)about 5.7% loss vs holding
Price ratio moves 5x25.5% loss vs holding

The loss grows faster than the price move. Fees can cover a small divergence but not a large one, which is why Chainlink points newcomers toward closely correlated pairs such as two stablecoins.

Approving the wrong token, or forgetting the approval. Before a contract can move a token from a wallet, the wallet has to grant it permission. Blockchain Council lists approving the wrong token, or skipping the approval entirely, as a common first error. The sign is a transaction that fails or a permission that is broader than intended. Phishing accounted for about 11% of third-quarter 2026 incidents, so an approval signed on the wrong site is the costlier version of the same mistake. Practices for checking and revoking approvals belong to the step after this one.

After the first read

The second thing is to repeat the exercise in a different category and compare the two paragraphs. Lending and liquid staking are the obvious pair, since they are among the largest categories and they depend on each other. The comparison shows something the first read cannot: a deposit in one protocol carries the risks of the protocols it rests on.

The second read should add one question the first did not ask: is a person making decisions with the money? SEC Commissioner Hester Peirce warned on July 24, 2026 that automated vaults and onchain lending products, a sector of about $131 billion, may be unregistered investment companies where human curators rebalance them at their discretion. Morpho's governance token fell sharply after the warning. Whether a protocol runs purely on fixed rules or on someone's judgment now matters for its legal standing as well as its risk.

Only after that does it make sense to deposit, and then in an amount small enough to learn the mechanics without the lesson being expensive.

Where the yield comes from, category by category

The beginner's guide to yield farming covers liquidity provision, lending yields and staking rewards, and explains how APR and APY are calculated.

Read the yield farming guide

Questions at this stage

It means a set of smart contracts on a blockchain that provides a financial service, such as trading, lending or staking, without a bank or broker in between. The Congressional Research Service adds that a truly decentralized protocol is autonomous, permissionless and transparent, with no central controlling party. Many protocols in practice still keep admin keys.

The main types are decentralized exchanges, lending markets, liquid staking, restaking, stablecoins, bridges and derivatives. On DefiLlama's late-September 2026 figures, the largest single protocol was Lido (liquid staking, about $26.46 billion) and the largest lending protocol was Aave (about $19.145 billion).

Not by default. September 2026 was the worst month of the year, with $766.49 million stolen across 99 incidents. The losses came mainly from compromised keys, bridge and contract logic flaws, and price manipulation. Market risks such as liquidation and impermanent loss come on top of that. The risk differs a great deal between protocols, which is why each one has to be read on its own record.

No. Drift and KelpDAO were both audited before losses of $285 million and $290 million in April 2026. Both attacks went through compromised keys rather than flaws in the audited code, and together they accounted for about 44% of the year's losses through early September.

Increasingly, and unevenly. In the U.S., the SEC has warned that some vaults may be securities, and DEX advocates proposed an exchange-registration safe harbor on September 14, 2026. In the UK, the FCA's new crypto regime opens for authorization applications on September 30, 2026 and launches on October 25, 2027. A protocol's regulatory position can change after a deposit is made, so it belongs in the same note as its incident record.