Testing for Captcha (OWASP-AT-008)

Brief Summary
CAPTCHA ("Completely Automated Public Turing test to tell Computers and Humans Apart") is a type of challenge-response test used by many web applications to ensure that the response is not generated by a computer. CAPTCHA implementations are often vulnerable to various kinds of attacks even if the generated CAPTCHA is unbreakable. This section will help you to identify these kinds of attacks.

Description of the Issue
Although CAPTCHA is not an authentication control, its use can be very efficient against:


 * enumeration attacks (login, registration or password reset forms are often vulnerable to enumeration attacks - without CAPTCHA the attacker can gain valid usernames, phone numbers or any other sensitive information in a short time)
 * automated sending of many GET/POST requests in a short time where it is undesirable (e.g., SMS/MMS/email flooding), CAPTCHA provides a rate limiting function
 * automated creation/using of the account that should be used only by humans (e.g., creating webmail accounts, stop spamming)
 * automated posting to blogs, forums and wikis, whether as a result of commercial promotion, or harassment and vandalism
 * any automated attacks that massively gain or misuse sensitive information from the application

Using CAPTCHAs as a CSRF protection is not recommended (because there are stronger CSRF countermeasures).

These vulnerabilities are quite common in many CAPTCHA implementations:


 * generated image CAPTCHA is weak, this can be identified (without any complex computer recognition systems) only by a simple comparison with already broken CAPTCHAs
 * generated CAPTCHA questions have a very limited set of possible answers


 * the value of decoded CAPTCHA is sent by the client (as a GET parameter or as a hidden field of POST form). This value is often:
 * encrypted by simple algorithm and can be easily decrypted by observing of multiple decoded CAPTCHA values
 * hashed by a weak hash function (e.g., MD5) that can be broken using a rainbow table
 * possibility of replay attacks:
 * the application does not keep track of what ID of CAPTCHA image is sent to the user. Therefore, the attacker can simply obtain an appropriate CAPTCHA image and its ID, solve it, and send the value of the decoded CAPTCHA with its corresponding ID (the ID of a CAPTCHA could be a hash of the decoded CAPTCHA or any unique identifier)
 * the application does not destroy the session when the correct phrase is entered - by reusing the session ID of a known CAPTCHA it is possible to bypass CAPTCHA protected page

Black Box testing and example
Use an intercepting fault injection proxy (e.g., WebScarab) to:

Find out if similar CAPTCHAs have already been broken. Broken CAPTCHA images can be found here gimpy, PWNtcha, and lafdc.
 * identify all parameters that are sent in addition to the decoded CAPTCHA value from the client to the server (these parameters can contain encrypted or hashed values of decoded CAPTCHA and CAPTCHA ID number)
 * try to send an old decoded CAPTCHA value with an old CAPTCHA ID (if the application accepts them, it is vulnerable to replay attacks)
 * try to send an old decoded CAPTCHA value with an old session ID (if the application accepts them, it is vulnerable to replay attacks)

Verify if the set of possible answers for a CAPTCHA is limited and can be easily determined.

Gray Box testing and example
Audit the application source code in order to reveal:


 * used CAPTCHA implementation and version - there are many known vulnerabilities in widely used CAPTCHA implementations, see http://osvdb.org/search?request=captcha
 * if the application sends encrypted or hashed value from the client (which is a very bad security practice) verify if used encryption or hash algorithm is sufficiently strong