In the first six months of 2026, attackers took more than $1.3 billion from crypto users and protocols across 344 separate incidents, and the majority of that money was lost through wallets, signatures and stolen keys rather than broken code. That number matters to anyone who swaps tokens on-chain, because a decentralized exchange never holds your assets: your wallet does, and every safeguard that a bank or a centralized platform would run for you becomes your own job the moment you connect. CertiK counted 204 code-level exploits in that same period, yet they cost only $151.6 million combined, while four social engineering operations alone drained $310 million, which tells you exactly where the real danger sits. This guide walks through the practical layers of DEX security, from wallet setup and contract checks to token approvals, slippage settings, trade routing and long-term habits, so you can trade on-chain without handing your balance to a phishing site. Let’s start with why security deserves the first slot in your trading routine, not the last.
A decentralized exchange settles trades through smart contracts, so there is no support desk to reverse a transfer and no account freeze to buy you time. Once a signature is broadcast and included in a block, the outcome is permanent. TRM Labs recorded 207 hacking incidents in the first half of 2026, the highest count it has ever logged for a six-month window, and more than $620 million of the stolen value was never recovered. The upside of permissionless trading is that nobody can block you. The downside is the same sentence read backwards: nobody can help you either. Security is what converts that trade-off into an acceptable one.
Risk on a DEX arrives from several directions at once. Your device or browser can be compromised. The front end you load can be a clone hosted on a look-alike domain. The token you buy can carry a contract that blocks selling. The pool you trade in can be too thin to absorb your order. The router you use can leak your transaction to bots that reorder blocks for profit. And the protocol itself can hold a bug that nobody found during review. Each of these has its own defence, which is why a single precaution, such as buying a hardware wallet, is never a complete answer.
On a centralized exchange, the platform holds the keys, runs the anti-fraud checks, and absorbs some categories of loss. In self-custody, all three duties transfer to you. That includes storing a recovery phrase where fire, theft and curious relatives cannot reach it, judging whether a contract deserves an allowance, and reading a signature request before approving it. The FBI’s Internet Crime Complaint Center logged 181,565 crypto-related complaints in 2025 with reported losses of $11.37 billion and an average loss per victim of $62,604, and the largest share of those cases came down to people being convinced to act, not to cryptography failing.
The two models fail in different ways, so they call for different habits. A centralized venue concentrates risk: when Bybit lost roughly $1.4 to $1.5 billion in February 2025, individual users were exposed to a breach of infrastructure they could not inspect. A DEX distributes risk: no single hack takes everyone’s balance, but every user carries their own attack surface. Custodial platforms can also freeze stolen funds, which on-chain venues generally cannot. Neither model is strictly safer. What changes is who makes the mistakes and who pays for them.
| Security aspect | Centralized exchange | Decentralized exchange |
|---|---|---|
| Who holds the assets | The platform | Your wallet |
| Main failure mode | Breach of the operator, insider access, withdrawal freeze | User signature errors, phishing, contract bugs |
| Loss recovery | Sometimes possible through freezes, insurance funds, courts | Rare, usually only through voluntary returns or whitelist freezes on stablecoins |
| Identity checks | Required in most jurisdictions | Not required at the protocol level |
| Counterparty for a trade | Internal order book run by the operator | Liquidity pool or solver network, settled by contract |
| Where you should focus | Account access, withdrawal whitelists, device hygiene | Key storage, signature review, approval management |
Traders who avoid losses tend to share one habit: they treat every signature as a payment. Before approving anything they ask what the transaction actually does, which contract receives permission, how much it can move, and whether the site asking for it was reached through a saved bookmark. That takes about fifteen seconds. Compare that against the median incident size CertiK reported for the first half of 2026, which was $138,703, and the fifteen seconds look cheap. Speed is what attackers rely on, so slowing down is itself a defence.
Self-custody means the private key that controls your assets exists only on hardware you own. No company can move those assets, and no company can stop you from moving them. In practical terms your balance is a set of entries on a public ledger, and the key is the only thing that authorizes changes to those entries. This is the base layer of every other tip in this guide. Wallet choice, approval hygiene and contract checks all exist to keep that key, and the signatures it produces, under your control.
A non-custodial wallet does not store your coins. It stores a key, reads balances from a node, builds transactions, signs them locally and broadcasts them. When you connect to a DEX, the site receives your public address and the right to propose transactions. It never receives permission to move anything by itself. Every transfer still needs a signature from your side. That design is what makes signature review the single most important skill in DeFi, because the wallet will happily sign whatever you approve, including a request that hands a stranger your entire token balance.
The private key controls one account. The seed phrase, usually 12 or 24 words, is the master input from which a wallet derives many private keys across many chains. Anyone holding the seed holds every account it generated, on every network, forever. That is why wallet software never needs it after setup, why no legitimate support agent ever asks for it, and why typing it into a website is equivalent to emptying your own accounts. Store it offline on paper or metal, keep at least one geographically separate copy, and never photograph it or save it to cloud storage.
The benefits are direct: no withdrawal limits, no account freezes, no exposure to an operator’s balance sheet, and access to protocols that no custodian lists. The risks are equally direct: no reset link, no fraud department, and no second chance after a bad signature. Self-custody also concentrates operational risk in your daily habits, from the browser extensions you install to the laptop you use on public networks. Traders who accept that trade-off deliberately, and build routines around it, generally do fine. Traders who drift into it without preparation are the ones filling incident reports.
The recurring errors are unglamorous. Storing a seed phrase in a notes app, a screenshot or a password manager that syncs to a cloud account. Reusing one wallet for airdrops, memecoins, long-term holdings and payments. Approving unlimited spending on a contract you tried once. Signing off-chain messages without reading them, which is how permit-based drains succeed without any on-chain approval appearing in your history. Clicking a link from a direct message. Each of these has cost people six-figure sums, and none of them require an attacker with advanced skills.
A hot wallet keeps keys on an internet-connected device, usually as a browser extension or a mobile app. It is fast, convenient and exposed to anything that compromises the device: malicious extensions, clipboard hijackers, info-stealer malware. A cold wallet keeps keys on a device that never exposes them, and signs transactions in isolation. The practical answer for active traders is not one or the other. Keep a small, refillable hot wallet for daily swaps, and hold the bulk of the portfolio in cold storage that only touches well-known protocols.
A hardware wallet protects the key, but its real value in DeFi is the screen. Because the device shows what you are actually signing, it gives you a second, independent view that malware on your computer cannot rewrite. That only helps if the device supports clear signing for the contract you are calling. Some interactions still display as raw data, and blind approval of unreadable payloads has produced serious losses, including the Bybit incident, where signers approved a transaction whose real effect was hidden from them. Choose a device with strong contract-decoding support and pair it with a wallet interface that simulates transactions before you confirm.
Splitting funds across dedicated wallets limits the damage of any single mistake. A common structure uses four: a vault for long-term holdings that never connects to a dApp, a main trading wallet for established protocols, a burner for new tokens and untested sites, and an airdrop or minting wallet for anything that requires unusual permissions. If a burner gets drained, you lose what was inside it and nothing else. This costs a little convenience and removes most of the tail risk from experimenting.
| Wallet type | Best use | Main strength | Main weakness |
|---|---|---|---|
| Browser or mobile hot wallet | Daily swaps, small balances | Fast, works with every dApp | Fully exposed to device compromise |
| Hardware wallet | Core holdings, larger trades | Keys never leave the device, independent confirmation screen | Blind signing risk when payloads are not decoded |
| Smart contract wallet or multisig | Shared treasuries, high-value storage | Spending rules, multiple approvers, social recovery | More complex, contract risk, higher gas costs |
| Burner wallet | New tokens, mints, unaudited sites | Losses are capped at the balance inside | Needs regular funding and clean-up |
Treat the seed phrase as the crown jewels and everything else as replaceable. Write it by hand, consider a metal backup for fire and water resistance, and split copies between two secure locations. Add a passphrase, sometimes called the 25th word, if your wallet supports it, and remember that losing it means losing access. Use a strong, unique password on the wallet application itself, enable a device PIN, and keep the machine you trade from free of pirated software. If you ever suspect exposure, move funds to a freshly generated wallet rather than hoping the exposure was harmless.
Label your wallets clearly so you never sign from the wrong one. Keep a written record of which protocols each wallet has interacted with. Rotate burner wallets every few months. Verify a receiving address by checking the first and last several characters, and send a small test transfer before a large one, which defeats address-poisoning attacks that seed your history with look-alike addresses. Finally, review balances and approvals on a schedule rather than only when something feels wrong.
Token tickers are not unique, and anyone can deploy a contract called USDC. Before trading a token, copy its contract address from a source you trust, such as the project’s official documentation, a major aggregator listing, or a block explorer entry with a verified label, and paste it into the DEX yourself. Do not rely on search results inside a swap interface, where fake entries appear constantly. Cross-check the address against two independent sources. On networks with many bridged assets, also confirm that you are looking at the correct chain’s version.
A block explorer tells you a lot in under a minute. Look at whether the source code is verified, how old the contract is, how many holders it has, and whether the deployer still controls functions that can mint tokens, pause transfers or change fees. Check the liquidity: is it locked, and for how long? Look at the largest holders, since a single wallet holding a dominant share is a warning. CertiK noted that a growing share of 2026 exploits hit contracts more than a year old, so age alone is reassurance, not proof.
An audit is an expert review of code against known failure patterns. It reduces risk, and its absence is a strong negative signal, but it is not a guarantee. Immunefi’s March 2026 analysis of 425 publicly known hacks between 2021 and 2025, covering $11.9 billion in losses, concluded that more than 90% of crypto projects still carry critical, exploitable vulnerabilities. Audits also go stale: code changes, dependencies change, and market conditions change. Prefer protocols with several audits from recognized firms, recent review dates, and a public record of how findings were fixed.
You do not need to read the code to get value from a report. Check five things. First, the scope: which contracts and which commit hash were reviewed, and whether the deployed version matches. Second, the date. Third, the severity table, especially critical and high findings. Fourth, the resolution status of each finding, since an acknowledged issue is not a fixed issue. Fifth, the centralization section, which usually describes what the team’s admin keys can do. That last part often reveals more real risk than the code findings themselves.
Warning signs cluster. Unverified source code. Functions that let an owner blacklist addresses or change transfer fees after launch. A proxy that can be upgraded by a single externally owned account. Extremely low or unlocked liquidity. Trading that works on the buy side while sells revert, which is the classic honeypot pattern. Token distribution where a handful of wallets hold most of the supply. Free contract scanners will flag many of these automatically, and when two or three appear together, the sensible response is to skip the trade.
Crypto phishing rarely tries to break encryption. It tries to get a valid signature from a legitimate owner. The attacker builds a convincing page, drives traffic to it through paid search ads, hacked social accounts, direct messages or fake support channels, and presents a transaction that looks routine. CertiK’s data shows how effective the targeted version has become: phishing incidents in the first half of 2026 fell to 63 from 132 a year earlier, but four operations produced $310 million in losses, roughly 85% of the phishing total, including one January case where a single victim lost about $284.8 million.
Clone sites copy the real interface pixel for pixel and sit on domains that differ by one character or use a different top-level domain. Some are promoted above the genuine site in search ads. Others appear as pop-ups claiming your wallet needs validation, synchronization or a security upgrade. Defences are simple and effective: reach exchanges only through your own bookmarks, verify the domain letter by letter before connecting, never enter a seed phrase into any web page for any reason, and treat any unexpected wallet prompt as hostile until proven otherwise.
The most common on-chain drain is not a hack at all. It is an approval you granted. A site asks you to approve a token before your first swap, which is normal, but the spender is an attacker-controlled contract and the amount is unlimited. Nothing happens immediately. The drain arrives later, once the balance is worth taking. A related version uses off-chain signatures such as Permit and Permit2, which produce no on-chain record at signing time, so nothing appears in your transaction history until the funds move.
The human layer is now the main target. Fake recruiters send take-home coding tasks that install info-stealers. Fake support agents appear in project chats within seconds of a public complaint. Fake investors invite founders to video calls with modified meeting apps. Romance and investment scams, which the FBI groups under relationship-driven investment fraud, accounted for $7.2 billion of reported US losses in 2025 across 61,559 complaints. AI has made all of this cheaper: the IC3 logged 22,364 complaints referencing AI-assisted fraud with $893 million in losses in its first year of tracking that category.
Check whether the team is public and whether their history can be verified outside the project’s own website. Read the documentation for specifics rather than promises. Confirm that audit links resolve to real reports on the auditor’s own domain. Look at the token distribution, the vesting schedule and whether liquidity is locked. Be suspicious of guaranteed returns, countdown timers, pressure to act before a deadline, and communities where questions get deleted. If a project cannot survive fifteen minutes of scrutiny, it will not survive your capital.
Under the ERC-20 standard and its equivalents, a contract cannot take tokens from your wallet unless you first grant an allowance. When you swap on a DEX, you approve the router to spend a specific token, and that permission stays active until you change it. Allowances are useful because they let a contract complete a trade without asking for a signature at every step. They are dangerous because they persist, and because the amount is frequently set far higher than the trade that prompted it.
Many interfaces default to an effectively unlimited allowance so users do not have to approve again on every trade. The trade-off is that if that contract is ever exploited or was malicious from the start, it can move your entire balance of that token, at any time, without further input from you. Several large drains have worked precisely this way, using allowances granted months earlier. Approving only the amount you intend to trade costs a little more gas and removes the exposure almost entirely.
Revocation sets an allowance back to zero. Tools such as Revoke.cash, Etherscan’s token approval checker and the built-in approval managers in modern wallets list every active allowance per address and let you cancel them in a few clicks. Do this whenever you stop using a protocol, after interacting with anything unfamiliar, and on a fixed schedule such as the first weekend of each month. One important limitation: an EIP-7702 delegation is not an ERC-20 approval and will not appear in standard revoke tools, which is covered below.
Set up passive monitoring so you are not relying on memory. Wallet interfaces such as Rabby and Frame show simulated outcomes and existing permissions before you confirm. Portfolio trackers and alert services can notify you when a new approval is granted or when a large transfer leaves your address. Explorers let you subscribe to address activity by email. The goal is to shorten the gap between an unexpected permission being granted and you finding out about it, because that gap is where losses grow.
Use exact-amount approvals for anything above a token amount you would be uncomfortable losing. Keep the number of protocols any one wallet has touched small. Revoke on a schedule, not on suspicion. Before signing an off-chain permit, check the spender, the value and the deadline shown in the wallet prompt, because a signature with a distant deadline is a standing invitation. After any suspected compromise, revoking is not enough on its own: move the assets to a new wallet generated from a new seed.
Before confirming, check that both sides of the pair are the tokens you intended, using contract addresses rather than tickers. On chains where the same asset exists in several bridged forms, confirm you are trading the canonical version, since the wrong wrapper can be nearly impossible to sell. Check the pool as well as the tokens: fee tier, current depth, and whether the price the interface quotes matches an independent reference. A quote far away from the market rate usually means thin liquidity, not a bargain.
Stick to venues with long operating histories, large audited codebases, meaningful volume and active security programs. According to DefiLlama’s dashboard, roughly $525 billion in spot DEX volume settled across a trailing 30-day window in spring 2026, with the top five venues handling around 65% of it, so there is no shortage of deep, established options. New exchanges with unusual token incentives may be legitimate, but they carry both contract risk and liquidity risk that established venues have already survived.
Thin pools punish you twice. Your own order moves the price against you, and the resulting slippage attracts bots. Before trading, compare your order size to the pool’s depth. A useful rule of thumb is to keep any single swap below roughly 1% of the pool’s liquidity, and to break larger orders into several smaller ones spaced out in time. Low-liquidity markets are also where exit scams work best, because a small amount of removed liquidity can make a token effectively unsellable.
Slippage tolerance is the worst price you agree to accept. Set it too tight and transactions fail; set it too loose and you are authorizing bots to take the difference. Research from Cointelegraph and EigenPhi covering more than 95,000 sandwich attacks estimated around $60 million in annual trader losses on Ethereum, with almost 40% of those attacks hitting low-volatility pools where users assumed they were safe. Individual sandwiches typically cost between 0.3% and 0.8% of trade value.
| Pair type | Suggested slippage | Why |
|---|---|---|
| Major stablecoin pairs | 0.05% to 0.1% | Deep liquidity, minimal genuine price movement |
| Blue-chip pairs such as ETH or BTC against a stablecoin | 0.1% to 0.5% | Deep pools, but real volatility during news |
| Mid-cap tokens | 0.5% to 1% | Thinner books, wider spreads |
| New or low-liquidity tokens | 1% to 3%, with a smaller order size | High impact per trade, high bot attention |
| Tokens with transfer fees | Fee plus a small buffer | The fee itself consumes part of the output |
Read the confirmation screen every time. Confirm the contract you are calling, the tokens leaving your wallet, the minimum amount you will receive, the network, and the gas cost. Use a wallet that simulates the transaction and shows the expected balance changes in plain language. If the simulation cannot resolve what the transaction does, or shows an approval you did not request, reject it. A rejected transaction costs nothing. A wrong signature is final.
An aggregator maintains integrations with vetted pools and market makers, then routes your order to the best available combination. In practice that means you interact with a reviewed router instead of hunting for pools yourself, which cuts the chance of landing in a malicious or abandoned pool. Good aggregators also flag or exclude tokens with unusual transfer logic. That does not remove the need to verify the token yourself, but it removes a whole class of accidental exposure.
Routing engines split an order across several pools and sometimes several chains, so no single pool absorbs the full size. Splitting a large swap into fractions across four or five venues can cut price impact substantially compared with executing it in one pool. The saving grows with trade size, which is exactly where manual execution performs worst. For traders moving meaningful size, routing quality is not a convenience feature, it is a direct reduction in loss.
Intent-based systems take this further. Instead of broadcasting a transaction that anyone can see and reorder, you sign an intent describing the result you want, and professional solvers compete to fill it. The order does not sit in the public mempool waiting to be sandwiched. Batch auction designs, private transaction relays and RFQ-style execution all reduce the same exposure. For large orders, this is currently the most reliable protection against transaction-ordering attacks.
The quoted rate on a single venue is rarely the best rate available. Aggregators poll many sources at once and account for gas, fees and price impact, which changes the ranking on smaller trades where gas dominates. Comparing quotes across two aggregators before a large trade takes seconds and regularly reveals a better route. Over a year of active trading, the difference compounds into a meaningful figure without any additional risk.
A route can look attractive in headline terms and still be poor. Multi-hop paths add gas and introduce extra pools, each with its own depth and fee. Routes through obscure venues add contract risk that is not reflected in the price. Check how many hops a route involves, which protocols it touches, and whether the final amount after gas actually beats a simpler alternative. If an unusually good quote depends on a venue you have never heard of, treat the quote as a warning rather than an opportunity.
Prefer protocols that have been reviewed by more than one reputable firm, that publish the reports, and that keep them current with the deployed code. Give extra weight to systems that have operated for years under real load with meaningful value at stake, because time under adversarial conditions is a form of testing no report can replace. Where a protocol is new, size your position to what you can lose, regardless of how impressive the audit list looks.
Look at how a team behaves, not only at what it publishes. Are contract upgrades controlled by a multisig or by a single key? Is there a timelock on changes, giving users a window to exit? Are oracle sources diversified? Is there an emergency pause, and who can trigger it? The Venus Protocol case in September 2025 is a useful benchmark: monitoring detected suspicious activity roughly 18 hours before the attack, and funds were recovered within hours. That outcome came from operational readiness, not luck.
Bug bounties pay researchers to find problems before criminals do. Immunefi reports that it protects more than $190 billion in value across over 230 active programs, and its payouts in the first quarter of 2026 reached $7.87 million across 1,104 paid reports, with the average reward per report roughly tripling from the previous quarter. A large, well-funded bounty with clear scope and a track record of payments is a meaningful signal that a team takes security seriously and can absorb bad news without shooting the messenger.
Some protocols maintain a reserve fund, funded by fees, that covers shortfalls after an exploit or a bad debt event. Third-party cover markets also sell protection against specific contract failures and stablecoin depegs. Neither is a full guarantee: reserves can be too small for a large incident, and cover policies have exclusions and claim processes. Still, a protocol that sets aside capital against failure is planning for the worst case, and users can size positions with that backstop in mind.
Risk is not static. Governance votes can change collateral factors, add new assets, alter fee routing or hand new powers to a committee. Contract upgrades can introduce bugs into code that was previously safe. Follow the governance forum and announcement channels of any protocol holding a significant share of your capital, and pay particular attention to proposals that touch admin permissions or oracle configuration. If you cannot follow the changes, keep the position small.
Leverage compresses the time between a bad decision and a total loss. On-chain perpetual venues can liquidate positions within a single volatile candle, and during network congestion your ability to add margin or close out may be limited exactly when you need it. Use leverage that survives a routine drawdown rather than the one you have seen recently, keep a margin buffer well above the liquidation threshold, and understand the funding rate you pay to hold the position.
Position size is the risk control that works even when everything else fails. Decide in advance the maximum share of your portfolio any single token, protocol or chain may represent, and apply the limit without exceptions. Spread exposure across uncorrelated assets and across venues, so one exploit cannot reach everything. Keep part of the portfolio in assets that do not depend on a single contract’s continued correctness. Diversification does not raise returns in a good month, but it keeps you solvent in a bad one.
Most self-inflicted losses in DeFi start with urgency. A launch that ends in an hour, a token pumping on a timeline, a message saying you need to act now. Attackers engineer this deliberately, and the FBI has noted that fraudsters manipulate emotion specifically to shorten the victim’s decision time. Write your rules while calm, define entry, exit and maximum loss before you open a position, and adopt one non-negotiable habit: never sign a transaction while you feel rushed.
Classic stop orders do not exist natively on most spot DEXs, but the function can be reproduced. Some perpetual venues offer on-chain conditional orders. Some aggregators and trading interfaces support limit and trigger orders through keeper networks. Where automation is unavailable, use price alerts and manual exits at predefined levels. Decide the invalidation point before you enter, and record it. A plan written in advance is what stops a bad trade turning into a permanent one.
Liquidity is not constant. A pool that easily absorbed your position last week can be a fraction of that size today, particularly after incentive programs end. Check pool depth before entering and before exiting, and remember that exit liquidity matters more than entry liquidity. During high volatility, spreads widen, price impact rises and failed transactions become more common. On chains where fees spike under load, the cost of exiting can itself become a factor worth planning around.
Route every meaningful transaction through a device that displays the details independently of your computer. Turn on any blind-signing restrictions the device offers, and only relax them when you genuinely need to interact with a contract the device cannot decode, then turn them back on. Combine this with a wallet interface that simulates transactions, so you see both a plain-language prediction of the outcome and an independent confirmation of the raw call.
Keep a vault wallet that never connects to any dApp. Assets there can only move if you initiate a transfer yourself, which means no approval, no delegation and no signature request can reach them. Fund the trading wallet from the vault in planned amounts. This one structural decision neutralizes most drainer attacks, because the compromised wallet simply does not contain the majority of your capital.
Beyond the vault, separate wallets by activity type and risk level. New protocols, token launches and anything requiring an unusual permission belong in a burner. Established lending and trading protocols can share a main wallet. Consider a separate browser profile, or a separate machine, for high-value operations, with no unrelated extensions installed. Browser extensions have direct access to page content and are a recurring source of compromise.
Set up alerts on every address that holds value: notifications for outgoing transfers, for new approvals and for large balance changes. Review approvals monthly. Check periodically whether your address carries an EIP-7702 delegation, since those do not appear in standard approval managers, and an unexpected delegation means the wallet should be treated as compromised rather than simply cleaned up. Early detection is often the difference between losing one token balance and losing everything.
Update wallet firmware, wallet applications, browsers and operating systems promptly, and always through the official channel rather than a link sent to you. Remove extensions you no longer use. Avoid installing wallet software from search advertisements, which have repeatedly hosted malicious clones. If you use a mobile wallet, keep the device free of side-loaded applications. Attackers exploit known, patched vulnerabilities far more often than they discover new ones.
No legitimate wallet, exchange, support agent, developer or auditor will ever need your recovery phrase. Every request for it is theft. The same applies to entering it into a website to claim an airdrop, verify a wallet, restore access or run a migration. If a phrase has ever been typed into a web page, photographed or stored in a cloud-synced file, treat it as public and move the assets to a new wallet immediately.
Connecting is not the same as approving, but it exposes your address and lets a site present prompts. Malicious pages use that to request approvals and delegations that look like routine steps. Connect only from bookmarked URLs, use a burner wallet for anything unfamiliar, and disconnect sites when you finish. Reviewing your list of connected applications every month takes a minute and closes doors you forgot were open.
Blind signing is the single most expensive habit in DeFi. Attackers depend on users approving payloads they have not read, which is how batched EIP-7702 phishing transactions succeed: one signature can authorize a delegation plus a series of approvals and transfers, all executed atomically. Wintermute’s research after the Pectra upgrade found that more than 97% of delegations on Ethereum mainnet at that point pointed to copy-pasted sweeper contracts, and individual victims have lost seven-figure sums to a single click.
Leverage is not a security failure, but it produces the same result. Positions sized for perfect conditions are liquidated by ordinary volatility, and on-chain liquidations happen without warning or discretion. Traders who survive tend to use modest leverage, hold spare margin, and accept smaller returns in exchange for staying in the game. There is no risk management technique that compensates for a position that cannot survive a normal week.
Yields on new protocols are high because the risk is high. An unaudited contract holding your deposit is an untested claim that the code does what the documentation says. Even audited code fails, which is why Immunefi’s dataset covering 425 hacks and $11.9 billion in losses is relevant here: audits reduce risk, they do not remove it. If you choose to farm on something new, treat the entire deposit as capital you can afford to lose, and do not let it grow into a position you cannot.
Detection systems now analyze transaction patterns, contract bytecode and front-end behaviour in real time, flagging drainer signatures and suspicious approvals before a user confirms. Wallets increasingly ship with built-in warning layers that simulate a transaction and label the counterparty. The same technology helps attackers, which is why the FBI began tracking AI-assisted fraud as a separate category in 2025 and recorded $893 million in losses in the first year. Expect the defensive side to move from optional plugins to default wallet behaviour.
Smart accounts allow spending limits, time locks, session keys, multi-factor approval and social recovery without a seed phrase. EIP-7702, delivered with the Pectra upgrade, lets an ordinary externally owned account take on smart-contract behaviour, which unlocks batching, gas sponsorship and recovery for wallets that already exist. The same feature created a new phishing surface, since a single delegation signature can hand execution rights to an attacker. The technology raises the ceiling on wallet security while raising the cost of one careless signature.
Clear signing, where a wallet decodes a transaction into plain language rather than showing raw data, is becoming a baseline expectation rather than a premium feature. Standards for readable transaction metadata, transaction simulation, phishing-site blocklists and delegation warnings are spreading across major wallets. As these become defaults, the burden on users falls, though the most important decision, whether to sign at all, will remain human.
Intent architectures replace “execute this exact transaction” with “achieve this outcome,” letting solvers compete off-chain and settle on-chain. Users get better prices, protection from public mempool exposure and fewer failed transactions. Combined with encrypted mempools and protocol-level ordering reforms, this steadily reduces the value attackers can extract from ordinary trades. Adoption is already visible in the routing choices of the largest aggregators.
The direction of travel is clear from the incident data. Code exploits are becoming more numerous but smaller, while infrastructure, key management and social engineering account for most of the money lost. That is pushing the industry toward continuous monitoring instead of one-off audits, formal verification for critical components, on-chain circuit breakers, coordinated freeze mechanisms with stablecoin issuers, and insurance markets that price protocol risk. Security is shifting from a launch checkbox to an operating discipline.
| Attack vector | Recent scale | Primary defence |
|---|---|---|
| Phishing and social engineering | About $310 million from four operations in H1 2026, roughly 85% of all phishing losses | Bookmarked URLs, signature review, hardware confirmation, burner wallets |
| Wallet and private key compromise | The largest driver of Q2 2026 losses at $807.5 million across all vectors | Hardware wallets, offline seed storage, wallet separation, clean devices |
| Malicious approvals and delegations | Over 97% of early EIP-7702 delegations pointed to sweeper contracts | Exact-amount approvals, monthly revocation, delegation checks |
| Smart contract exploits | 204 incidents in H1 2026 but only $151.6 million in losses | Audited protocols, position limits, timelocks and monitoring |
| Transaction ordering and MEV | Around $60 million a year from Ethereum sandwiches | Tight slippage, private relays, intent-based execution |
| Fake tokens and rug pulls | A persistent share of retail losses on new listings | Contract address verification, liquidity checks, holder analysis |
Security on decentralized exchanges is not a single product you buy, it is a set of habits you repeat. The data from 2025 and the first half of 2026 points in one direction: broken code costs users far less than broken judgement, and the money follows phishing, key compromise and careless approvals. Everything in this guide is built around that fact. Keep long-term holdings in cold storage that never touches a dApp. Trade from a small, funded hot wallet. Verify contract addresses instead of trusting tickers. Read every signature request before approving it. Set slippage deliberately and route larger orders through established aggregators. Revoke old approvals on a fixed schedule and check your addresses for unexpected delegations. Start today with the two steps that take the least time and prevent the most damage: revoke the approvals you no longer need, and move your long-term holdings to a wallet that never connects anywhere. Then trade with the rest, knowing that a single bad click can no longer cost you everything.
Every trade you make in decentralized finance is handled by a piece of...
Imagine placing a currency trade that no bank, broker, or middleman ever touches ...
Slippage is the quiet tax that eats into almost every trade you make...
Every time you swap one currency for another on-chain, a hidden engine sets...
Behind every instant currency swap on a decentralized exchange sits a shared pot...
Every trade you make in decentralized finance is handled by a piece of...
Imagine placing a currency trade that no bank, broker, or middleman ever touches ...
Slippage is the quiet tax that eats into almost every trade you make...
Every time you swap one currency for another on-chain, a hidden engine sets...
Behind every instant currency swap on a decentralized exchange sits a shared pot...
