Transaction authorization: confirming a sensitive action so it cannot be tampered with
Design separate confirmation for sensitive actions with server-side checks, significant transaction details and single-use authorization.
When an app lets someone move money, change payout details or approve a sensitive step, a login is not enough. OWASP's Transaction Authorization Cheat Sheet is about the extra confirmation step (it calls these actions "transaction authorizations") and how attackers get around it. We read it in full on 15 November 2026. We did not read its references or the ASVS requirement it cites. It is written for banks, developers and penetration testers, and we summarise it for product owners.
What it covers
OWASP says apps usually ask for a second factor to confirm a sensitive operation such as a wire transfer, and that the idea spreads beyond finance (an emailed unlock code or token link is a transaction authorization). Methods it lists: a card with authorization numbers, a time-based one-time password (TOTP) token, a code sent by SMS or phone, a digital signature from a smart card or smartphone, or a challenge-response token including card readers that are not connected to the computer.
What a user must see: "What You See Is What You Sign"
- The authorization method must let the user identify and acknowledge the significant data of the transaction, for a wire transfer the target account and amount. OWASP says developers cannot assume the user's computer is secure, so an external component should check the data. What counts as significant depends on real risk, the method's technical limits and the user experience. For methods where typing data is awkward, such as an unconnected card reader, return the minimum, for example a partial account number and the amount.
- Without a meaningful prompt, social engineering and malware can abuse the process.
- Authentication and authorization should look different. If sign-in and transaction confirmation use the same action (the same OTP token, for example), malware can show a false error after sign-in so the user repeats the step, and use the second code to authorize a fraudulent transaction. OWASP says even challenge-response can be abused this way, and that the attack is widely used against online banking malware. Fixes it lists: different methods for each, different modes in an external component, and a clear message about what the user is signing.
Rules for changing the authorization setup
- Changing the authorization token (such as the phone number for SMS codes) should be authorized with the current token, for example by sending the code to the current phone number.
- Changing the authorization method should be authorized with the current method, so malware cannot downgrade the user to the weakest method; tell users about the dangers of their chosen method.
Server-side rules (non-functional guidelines)
- Enforce on the server. Altering client data (changing parameters, adding or removing parameters, causing an error) must not change the outcome. OWASP lists default deny and no debugging features in production.
- Enforce the method on the server, so a client cannot downgrade it. When adding a new, stronger method on top of an old codebase, attackers may still authorize by sending the old method's parameters.
- Generate verification data on the server. Significant transaction data should be generated and stored on the server and passed to the authorization component without client tampering; data collected on the client could be altered by malware to show fake data.
- Stop brute force. After a set number of failed attempts, restart the whole authorization.
- Control state transitions. The flow (enter data, request authorization, initialise, user confirms data, user responds, server validates and executes) must run in order with no skipping or overwriting.
- Protect transaction data from change after the user entered it. OWASP's example is a time-of-check to time-of-use attack: authorize one transaction, execute another. If data changes, invalidate any entered authorization and reset the process; attempts to change data after entry should be logged, monitored and investigated.
- Keep transaction data confidential in client-server traffic.
- A final gate before execution that checks the transaction was properly authorized.
- Time-limit credentials, since malware may pass codes to an attacker who uses them from another machine. Choose the window so it does not disrupt normal use.
- Make credentials unique per operation (a timestamp, sequence number or random value in the signed data or challenge). The sheet adds that each credential must be accepted only once, with an atomic check that it is unused and a mark that it is now used, using state shared by all servers that accept it.
Beyond the sheet's scope (its own remarks)
Which transactions need authorization is a risk decision for the app owner. OWASP recommends cryptographic protection for integrity, confidentiality and non-repudiation, careful provisioning and protection of device signing keys (malware may inject, replace or steal them), user awareness (copy the data from a trusted source, not from the computer screen), and notes anti-malware helps but cannot be fully effective.
Questions for your team
- Which actions need a separate confirmation, and who decided?
- Does the user see the amount and destination in the confirmation step itself?
- Is the confirmation different from sign-in?
- Can any step be skipped, repeated or altered by changing a request?
- What happens when someone changes their phone number or method?
Related: MFA options compared, forgot-password flows, session management basics.
Limits
One OWASP page, read in full. The sheet's references, the ASVS requirement and the cited conference talks were not read. No numbers for time windows or attempt limits are given by the source and we give none. It makes no claim about any product. Regulated finance products have legal requirements outside this article; talk to counsel.
By Omnistu.