Case study

Wallet drainers: How token approvals expose funds

The 2023 Ledger Connect Kit incident illustrates signing risk. See what a token allowance permits, what it leaves outside its scope and why disconnecting is different.

Dylan Foster Updated October 21, 2026 20 min read

Wallet drainers: How token approvals expose funds
Illustrative photograph. Photo by Tima Miroshnichenko. Pexels. Pexels License.
In this case study · 12 sections

A wallet drainer can exploit an asset permission you signed rather than steal your recovery phrase. MetaMask explains that token approvals let an application access and move a particular token. A fraudulent request abuses that legitimate mechanism. Keeping recovery words private remains necessary, but it does not make every wallet prompt safe.

The Ledger Connect Kit incident on 14 December 2023 provides the historical case for this explanation. The sections below separate the incident record from the permissions a reader might encounter now: an ordinary token allowance, a signed permit, an NFT operator approval or a direct transfer. These are different actions, even when a website presents them through similar buttons.

The practical aim is to make an unfamiliar request easier to describe before you accept it, and to make an existing transaction easier to discuss with verified support. Start with the asset and the action, then establish the permission’s scope. The technical examples concern Ethereum standards and related applications. They are not a description of every blockchain, wallet or recovery method.

The Ledger Connect Kit incident shows the signing risk

Ledger’s report, published on 20 December 2023, describes publishing access obtained through phishing of a former employee. On 14 December, the attacker used that access to distribute malicious versions of Connect Kit. Applications loading the affected package could then present harmful signing requests.

Ledger says ecosystem partners alerted it to the incident and a corrected package was deployed. Cached copies took longer to clear. The report also records coordination with WalletConnect to disable the rogue connection project. These are the manufacturer’s findings about that incident, not an independent audit by this publication.

The reported requests varied by asset: approvals or permit messages for tokens, and transactions moving native assets. The account concerns misleading signatures through affected applications, rather than extraction of private keys from Ledger hardware. It does not establish that every Ledger device or every application was affected.

This is a historical case, not a new 2026 incident. The report’s proposed future safeguards are not evidence here that every proposed change was later implemented.

What does this case establish, and what remains a separate question?

The useful editorial lesson is to assess the action presented for signature independently of the name on the surrounding interface. A brand can help you identify the documentation you need, but it cannot answer every question about a particular request. That is why this case study turns from the incident to the permission itself rather than treating brand recognition as the final check.

Keep three kinds of statement separate when reading a security report: what the author says happened, how the relevant mechanism works, and what you should do in your own situation. A historical report may support the first. A technical specification can support the second. Neither automatically diagnoses a different transaction you are looking at today.

Avoid extending the incident beyond its evidence

Do not use the case as proof that a recent loss involved the same package, attacker or technique. To make that connection, an investigator would need evidence from the actual interaction. A familiar symptom, such as tokens leaving after a signature, is a starting question rather than a complete attribution.

The same discipline applies to an apparent absence of loss. If you did not encounter the documented incident, that does not settle the meaning of an unrelated approval you granted. Keep the question narrow: which permission exists for this asset, account and recipient? The answer should come from the relevant request and records, not from whether a historic headline sounds familiar.

Separate device trust from transaction intent

When examining a prompt, ask both where it is displayed and what it authorizes. These questions serve different purposes. A display you trust still needs an instruction you understand. Conversely, a reasonable sounding description on a website does not explain a different action shown by the wallet.

This is an approach to reading the evidence, not a claim that every reader can audit a smart contract. If the information is insufficient, preserve the uncertainty. Declining an unexplained request is a complete decision; you do not need to identify the exact malicious technique before stopping.

What an ERC20 allowance actually permits

An ERC20 approval gives a named spender permission for a particular token and amount. In the ERC20 specification, approve sets that value; allowance reports what remains. It does not expose recovery words or automatically authorize unrelated assets.

A very large allowance is still an amount. MetaMask’s approval guide shows how spending caps can exceed the balance currently held. The distinction matters: limiting an allowance reduces the authorized amount, but it does not make a malicious spender trustworthy.

ERC20 approval

Scope: Spending allowance for a named token and spender.

Check: Token contract, spender and value.

Signed permit

Scope: A supported token’s allowance can be set using an authorized message.

Check: Token, spender, value and deadline.

Native transfer

Scope: Moves the native asset through the signed transaction.

Check: Destination, amount and transaction purpose.

Site connection

Scope: Access to account information and ability to request actions, depending on permissions.

Check: What was connected and what was separately signed.

These examples separate mechanisms; they are not a complete list of wallet capabilities. NFT permissions and other token standards require their own interpretation.

Read the permission as a complete statement

For your notes, express a request in a sentence that names the account, token, spender and amount. If you cannot fill one of those parts from reliable information, mark it as unknown. This is an editorial reading aid. It does not certify the contract or replace the provider’s technical instructions.

The benefit is that the missing detail becomes visible. “Allow the app” gives you very little to check. A description tied to the specific asset and recipient gives support a concrete question. Do not substitute a logo or short display name where you still need to identify the party receiving permission.

Keep the network alongside the statement. This article does not establish the same permission model for a different chain simply because an asset uses a similar name there. If the network or token standard differs from the example, consult the relevant documentation before carrying the conclusion across.

Worked example: balance, allowance and remaining exposure

This hypothetical example uses invented whole token amounts. It is a reading exercise, not a transaction to perform or an estimate of a real loss. Assume a conventional token with a finite allowance, no transfer fees and no other permissions or transfers. The spender is called Service A solely to keep the explanation readable; it does not identify a real service.

OpenZeppelin’s ERC20 reference describes spending against the caller’s allowance. For this example, we assume that a successful delegated transfer reduces the finite allowance by the amount transferred. Different implementations and additional features require their own review.

Start with the two separate amounts

Assume your account holds 300 example tokens and grants Service A an allowance of 80. The account balance is 300; the permission amount is 80. Those numbers answer different questions. The first describes the holding in our example. The second describes the specified grant to this spender.

Suppose Service A then uses 30 of that allowance. Under the stated assumptions, 300 minus 30 leaves a balance of 270, while 80 minus 30 leaves a remaining allowance of 50. A record showing only the new balance would not tell the whole permission story. You would still need the allowance result to describe what remains authorised.

Change one assumption at a time

If the original allowance were 20 instead, the proposed use of 30 would exceed that finite grant. This simple comparison helps explain why the cap matters. It does not tell you whether Service A is trustworthy, whether you intended any transfer or whether another permission exists elsewhere.

If a later review shows an allowance different from your calculation, do not force the real transaction into the example. Check whether another action occurred, whether you are examining the same account and token, and whether the implementation matches the assumptions. The exercise is designed to make those questions easier to ask, not to override the record.

Use the calculation to identify the unanswered question

Ask what you actually need to know next. Are you checking whether the original grant was broader than intended, how much was moved, or whether spending access remains? These are separate enquiries. Saving the input amounts beside the relevant transaction references helps a reviewer see which one you are making.

Do not turn the example into a prediction that only 80 tokens could ever be lost from a real wallet. The assumptions exclude other approvals, new signatures, different assets and exposed credentials. The arithmetic is deliberately narrow. Its useful result is a clear description of one permission, not a total risk estimate.

Why a message signature can also carry risk

A message signature can be more than a login acknowledgement. The ERC2612 permit standard allows a supported token’s spending allowance to be changed using a signed message. Its conditions include the owner, spender, value, deadline and a nonce used to prevent reuse.

The signature alone is not the same as the allowance already being recorded on the blockchain. It can authorize a later submission that sets the allowance if the standard’s conditions are met. Do not assume a prompt is harmless merely because it does not ask you to send a token immediately.

Not every message signature uses this standard. If you cannot identify what the request authorizes, reject it and consult verified wallet documentation. The button’s reassuring label is not a technical explanation.

Structured information still needs interpretation

EIP712 defines a format for structured data signing. It is intended to make the data available for inspection rather than present only an opaque string. The standard does not itself provide replay protection. That distinction helps explain why the message format and the rules of a specific authorisation must be considered separately.

Readable fields are an opportunity to check the request, not a positive verdict about it. Compare the displayed purpose with the task you intended. If you cannot explain why a permission is needed, a clearer screen has identified a question; it has not answered it for you.

OpenZeppelin’s permit guidance cautions that a permit expresses an allowance, not an intention to use it in a particular way. Do not substitute a website’s promised purpose for the scope actually signed. Ask what the authorisation allows even if the website says it will only use that authority for a narrow benefit.

Why NFTs and native assets need different checks

An ERC20 amount is not a universal model for every asset permission. The relevant standard determines what the request means. Identify the asset type before applying an allowance example, and keep uncertainty about an unfamiliar standard explicit.

An NFT approval can have a different scope

The ERC721 standard distinguishes approval for an individual NFT from approval of an operator for all of an owner’s NFTs within that contract. An operator grant should therefore not be read as an ERC20 quantity cap. Check the contract, operator and whether the request concerns one item or the broader operator permission.

ERC1155 uses a contract that can represent multiple token types and provides operator approval over the owner’s tokens in that contract. Its scope is not automatically every asset on every network. The fact that an interface says all is incomplete without knowing which contract and permission that word refers to.

When you ask for help, include the standard if you know it, but do not guess from a token’s picture. Preserve the contract and network information available through verified tools. A support question about an NFT operator is clearer than a general statement that you approved a coin.

A direct native asset transfer is another action

Ethereum’s transaction documentation describes signed instructions that can transfer ETH or interact with a contract. For a simple native asset transfer, the destination and value are central. That is different from granting an ERC20 spender a future allowance.

A zero native asset value also should not be treated as a complete description of a contract interaction. Ethereum transactions can include input data as well as a value. Ask what the contract action does; do not judge an unfamiliar request solely by the headline amount of native currency shown.

Keep the limits of this guide visible

The standards discussed here belong to the Ethereum ecosystem. This article does not determine the meaning of a Bitcoin, Solana or other network’s request. Use the relevant wallet and network documentation rather than translating an ERC20 rule by analogy. Similar interface wording does not establish identical technical behaviour.

Disconnecting and revoking do different jobs

Disconnecting a website does not cancel token allowances already granted to it. MetaMask explicitly separates disconnection from revocation. Closing a browser tab therefore does not establish that spending access has ended.

Revoking an onchain allowance requires a transaction. Use the provider’s verified instructions for the relevant network, token and spender, then check the resulting status. Removing one permission is not proof that every other approval, signed authorization or exposed credential is safe.

Revocation addresses remaining access. It does not reverse assets already transferred. MetaMask’s incident guidance also treats disclosed recovery words as a separate compromise and warns against adding fee funds when an automated sweeper may be present. Do not send more crypto into a suspected compromised wallet as an improvised fix.

Define what you expect a remedial step to change

Before following instructions to remove access, name the permission involved and the evidence you will use to check the result. Is the instruction about a site connection, an existing token allowance or a different kind of authorisation? If you cannot tell, ask verified support to clarify before signing something presented as a fix.

Keep a record of both the action and its status. Clicking a button is not the same statement as having a completed result you have checked. If the tool reports a pending or failed action, preserve that information rather than recording the problem as resolved. The relevant provider instructions should determine how completion is verified.

Do not broaden a specific result into a wallet wide assurance. You may be able to establish what happened to one permission while still having questions about other activity. Write those questions down separately so that a successful check does not erase an unresolved exposure from the record.

What can a transaction record tell you?

A transaction record is evidence about activity, not a complete explanation of the conversation that led to it. MetaMask explains how to inspect public account activity using the appropriate network’s explorer. Keep the network and identifier together so that another person can examine the same record.

Preserve the record and your observation separately

Record the identifier, the network, the asset and the result you are trying to understand. Then add a separate sentence describing what you remember doing. For example, you might remember pressing a button labelled claim while the transaction record needs interpretation. Your memory of the label is relevant context, but it should not replace the record’s details.

Keep timestamps in the form shown by the source, including the time zone if it is available. If you later convert a time for your own timeline, retain the original too. This is a record keeping recommendation that makes comparison easier; it is not a requirement to reconstruct every second before seeking help.

Do not confuse the permission event with the loss

Ask the reviewer which record establishes the authorisation and which establishes an asset movement. They may be different parts of the sequence. Do not label every visible entry as a separate loss or assume the first unfamiliar item explains everything that followed.

Similarly, keep a quoted market value separate from the number of tokens recorded. A changing currency estimate is not itself another transfer. If you need to report an amount, identify the asset and quantity and explain where any currency conversion came from. Avoid inventing a precise Canadian dollar loss from an unexplained balance display.

A destination is not a verified personal identity

Use the destination address as an identifier in your record. Do not attach a person’s name merely because a chat account claims to control it or an unrelated post mentions it. Ask what independent evidence supports an identity claim before repeating it publicly. The responsible reporting task is to preserve what you know, including the limits of attribution.

An explorer page also does not reproduce every message exchanged before signing. Save the relevant approach separately, without wallet secrets. The two records serve different purposes: one helps examine activity, while the other describes how the request was presented to you.

Missing information remains missing

If you cannot retrieve the original prompt safely, say so. Do not recreate it by accepting another request or returning to a suspicious application. An honest gap is preferable to a reconstruction that accidentally presents a later screen as the one you signed.

When several interactions occurred, list them in order as far as you can establish it. Distinguish a connection, a signature you remember and a transaction you can identify. The aim is to give verified support a usable starting point, not to claim that you have independently solved the incident.

What does a verified contract label establish?

Source verification answers a narrower question than whether a contract is safe to use. Ethereum’s verification documentation explains the comparison between published source code and deployed bytecode. It distinguishes that process from verifying the correctness of the contract’s behaviour. Access to readable code enables examination; it does not mean every user has completed that examination.

For a reader assessing an approval, the practical implication is to ask what a reassuring label actually covers. Does the claim concern source code, an identified company, a particular audit or the permission on your screen? Keep those statements separate. A positive answer in one category should not silently become a positive answer in all the others.

Match any assurance to the exact object being checked

If someone points to a technical report, ask which contract address and network it concerns. Compare that scope with the request you are trying to understand. If you cannot establish the connection, record the assurance as unconfirmed for your request. This is a document comparison task, not an instruction to decide the security of code you cannot evaluate.

Read the limits of the assurance as carefully as its headline. A statement may cover a particular version or a defined review scope. Do not omit those qualifications when asking support about it. If the person recommending the transaction cannot explain how the evidence relates to the current permission, leave the permission unresolved.

Avoid turning a warning search into a guarantee

A search can help you locate documentation or a report about an address. For your decision record, write what you found rather than declaring the absence of a result to be proof of safety. An unanswered search and a completed technical assessment are different statements.

This also works in the other direction. A comment accusing a similarly named project does not by itself identify your contract or establish what happened to your assets. Preserve the original reference and ask a qualified reviewer to examine the match. Careful identification protects the quality of your incident report while leaving the immediate decision simple: do not sign a request whose purpose and scope you still cannot explain.

What to examine before and after signing

Before approving, match the wallet request to the action you intended. Identify the asset, the party receiving permission and its amount or scope. A familiar site name does not explain an unexpected spender or a broader permission than the task requires.

If you have already signed something suspicious, preserve the network, transaction identifier, affected asset and the request details you can retrieve safely. Use verified support to distinguish an approval, a permit or an actual transfer. Do not sign another unexplained request offered as a repair.

For immediate containment and Canadian reporting, use what to do after a crypto scam. The wallet phishing guide covers fake support and recovery phrase traps. The Canadian scam overview helps assess an unfamiliar approach before interacting.

Before signing, compare intent with authority

Describe the intended task in plain language before reading an unexpected request. You may intend to exchange a specified amount, list one item or view eligibility information. Keep that intention beside the request, then identify the difference. This is more useful than asking whether the screen looks professional.

If the scope appears wider, ask why and look for an explanation in independently obtained documentation. Do not let the requesting page answer every question about itself. If the explanation remains unclear, decline the request. You do not need to make a small transaction to earn the right to ask what it authorises.

Avoid treating several prompts as one decision. After completing one step, read the next request on its own terms. Your reason for accepting an earlier connection does not automatically explain a later transfer or permission. If the workflow changes direction, pause and reassess the new action.

How should you organise an unresolved signing incident?

Organise the situation around what is known, what is suspected and what still needs interpretation. This is a practical editorial framework, not a substitute for professional investigation. It helps keep a fear about an entire wallet from obscuring a specific question someone can address.

Start with the observed action

Write down whether you connected a site, signed a message, confirmed a transaction or disclosed a secret. Include more than one category when necessary. Avoid replacing these descriptions with the single word hacked, because that word does not identify the action a provider needs to examine.

If you saw an unexplained balance change, state which asset and account you checked. Then distinguish that observation from your theory about its cause. You can report a suspected drainer without claiming to know the exact mechanism. That leaves room for the evidence to correct the initial interpretation.

Choose the source that can answer the question

Use wallet documentation for the meaning of the displayed request and the provider’s incident process. Use the relevant standard when you need to understand a permission’s defined scope. Use official reporting channels for the incident itself. None of these requires giving an unknown helper your recovery phrase.

Ask a narrow question when possible: which token and spender does this approval concern, was the transaction completed, or is the record a transfer rather than a permission change? A specific question is easier to evaluate than a promise that one service can clean everything. Keep the response tied to the evidence supplied.

Treat a proposed repair as another request

Do not suspend the normal checks because an instruction is described as urgent remediation. Identify the website, requested permission and expected result again. If the proposed repair asks for an unexplained signature, a new payment to a stranger or disclosure of recovery material, stop and return to verified support.

Keep the original incident and the new approach in separate notes. That makes it possible to explain whether the person offering help contacted you first and what they asked you to do. Do not describe an unverified recovery claim as a service you have already established to be legitimate.

Leave an explicit list of open questions

After a review, note what was actually confirmed and what remains unresolved. You might understand one allowance but still be unable to identify a separate signed message. State both. Do not erase the second question because the first received a reassuring answer.

This record also helps if you change support channels. The next person can see the scope of the earlier check rather than assume the whole wallet was assessed. Share only the evidence relevant to that channel and keep keys, recovery words and unrelated personal information out of the exchange.

Questions about wallet drainers

Can tokens be stolen without someone knowing my seed phrase?

Yes. A malicious spender can use a permission you granted within its scope. A signed transfer can also move assets without disclosing your phrase. The relevant question is what you authorized.

Does one approval expose my entire wallet?

Not automatically. The ERC20 example applies to a particular token contract, spender and allowance. Other permissions or signatures can create additional exposure, so inspect the actual requests rather than assuming one universal wallet permission.

Does an unlimited approval mean no amount was specified?

No. An ERC20 approval includes a value. An allowance described as unlimited is a large grant, not the absence of an amount. Check what the wallet displays for that token and spender.

Will revoking an approval return stolen tokens?

No. Revocation removes the relevant remaining spending permission when completed; it does not undo earlier transfers. Preserve the evidence and follow the incident response process rather than paying someone who promises a reversal.

Is a signature without a gas payment harmless?

No. The ERC2612 example shows why a signed authorisation can matter before the signer submits an onchain transaction. Whether a fee appeared is not enough to determine the meaning of the request. Identify the message type and its scope, and seek verified guidance if those details are unclear.

Does the permit deadline end an existing allowance?

In ERC2612, the deadline limits when the permit may be accepted. It is not an automatic expiry field for an allowance already established by a valid permit. Do not confuse the signing deadline with proof that spending access has ended. Check the actual current permission for the relevant token and spender.

Can a transaction marked successful still be a scam?

A completed technical action does not establish that the explanation given to you was honest. Keep the transaction result separate from the purpose you were told it served. When asking for help, provide both the record and the relevant message rather than expecting the completion label to decide the dispute.

Should I use another blockchain’s instructions to fix this?

Use guidance for the actual network, wallet and permission involved. This article’s ERC20 and permit examples do not define every network’s account model. If a tool or instruction does not support your situation, do not improvise by changing networks or approving something similar. Ask the provider to identify the applicable process.

Does a clean permission review prove no phrase was exposed?

No. A review of token permissions addresses that category of access. It does not establish what you typed into a website, sent in a message or exposed through software. Describe those actions separately and use the phishing guide if recovery material or account credentials were involved.

Sources