Pseudorandom number generators (PRNGs) use deterministic mathematical algorithms to produce a sequence of numbers with good statistical properties. However, the sequences of numbers produced fail to achieve true randomness. PRNGs usually start with an arithmetic seed value. The algorithm uses this seed to generate an output value and a new seed, which is used to generate the next value, and so on.

The Java API provides a PRNG, the java.util.Random class. This PRNG is portable and repeatable. Consequently, two instances of the java.util.Random class that are created using the same seed will generate identical sequences of numbers in all Java implementations. Seed values are often reused on application initialization or after every system reboot. In other cases, the seed is derived from the current time obtained from the system clock. An attacker can learn the value of the seed by performing some reconnaissance on the vulnerable target and can then build a lookup table for estimating future seed values.

Consequently, the java.util.Random class must not be used either for security-critical applications or for protecting sensitive data. Use a more secure random number generator, such as the class.

Noncompliant Code Example

This noncompliant code example uses the insecure java.util.Random class. This class produces an identical sequence of numbers for each given seed value; consequently, the sequence of numbers is predictable.

import java.util.Random;
// ...

Random number = new Random(123L);
for (int i = 0; i < 20; i++) {
  // Generate another random integer in the range [0, 20]
  int n = number.nextInt(21);

Compliant Solution

This compliant solution uses the class to produce high-quality random numbers:

// ...

public static void main (String args[]) {
  SecureRandom number = new SecureRandom();
  // Generate 20 integers 0..20
  for (int i = 0; i < 20; i++) {

Compliant Solution (Java 8)

This compliant solution uses the SecureRandom.getInstanceStrong() method, introduced in Java 8, to use a strong RNG algorithm, if one is available.

// ...

public static void main (String args[]) {
   try {
     SecureRandom number = SecureRandom.getInstanceStrong();
     // Generate 20 integers 0..20
     for (int i = 0; i < 20; i++) {
   } catch (NoSuchAlgorithmException nsae) { 
     // Forward to handler


MSC02-J-EX0: Using the default constructor for java.util.Random applies a seed value that is "very likely to be distinct from any other invocation of this constructor" [API 2014] and may improve security marginally. As a result, it may be used only for noncritical applications operating on nonsensitive data. Java's default seed uses the system's time in milliseconds. When used, explicit documentation of this exception is required.

import java.util.Random;
// ...

Random number = new Random(); // Used only for demo purposes
int n;
for (int i = 0; i < 20; i++) {
  // Reseed generator
  number = new Random();
  // Generate another random integer in the range [0, 20]
  n = number.nextInt(21);

For noncritical cases, such as adding some randomness to a game or unit testing, the use of class Random is acceptable. However, it is worth reiterating that the resulting low-entropy random numbers are insufficiently random to be used for more security-critical applications, such as cryptography.

MSC02-J-EX1: Predictable sequences of pseudorandom numbers are required in some cases, such as when running regression tests of program behavior. Use of the insecure java.util.Random class is permitted in such cases. However, security-related applications may invoke this exception only for testing purposes; this exception may not be applied in a production context.

Risk Assessment

Predictable random number sequences can weaken the security of critical applications such as cryptography.




Remediation Cost









Automated Detection



Hardcoded Random Seed (Java)
Insecure Random Number Generator (Java)
Risky Cipher Field (Java)
Risky Cryptographic Algorithm (Java)
Risky Cryptographic Field (Java)
Unsafe Base64 Encoding (Java)
Weak Hash Algorithm Field (Java)
Legacy Random Generator (Java)

Parasoft Jtest
CRT.MSC02.SRDUse '' instead of 'java.util.Random' or 'Math.random()'

Related Vulnerabilities

CVE-2006-6969 describes a vulnerability that enables attackers to guess session identifiers, bypass authentication requirements, and conduct cross-site request forgery attacks.

Related Guidelines

SEI CERT C Coding Standard

MSC30-C. Do not use the rand() function for generating pseudorandom numbers

SEI CERT C++ Coding Standard

MSC50-CPP. Do not use std::rand() for generating pseudorandom numbers


CWE-327, Use of a Broken or Risky Cryptographic Algorithm

CWE-330, Use of Insufficiently Random Values

CWE-332, Insufficient Entropy in PRNG

CWE-336, Same Seed in PRNG

CWE-337, Predictable Seed in PRNG



    • The 1st noncompliant code does indeed produce the same string of random numbers for each invocation. This is not always a Bad Thing. For instance, when performing regression tests of program behavior, you ideally want all invocations to use the same sequence of random numbers. It is important, as you note, that this behavior does not persist in production code. The example is valid, but the discussion should make this clear.
    • The 2nd noncompliant code produces unique random number sequences on each invocation, for me, probably because each RNG is seeded at a different time. I'm not sure this is a security hole, although the production of each RNG is inefficient. See Elpiniki's MSC32-C for a discussion on the merits of seeding RNGs with current time.
    • The compliant solution doesn't work 'out of the box'. I get:
    Exception in thread "main" SHA-1 SecureRandom not available
    	at Foo.main(

    At least, it needs to catch NoSuchAlgorithmException or declare it throws this exception.

    • Needs 'Risk Assessment' and 'Reference' sections.
    1. As far as the 2nd noncompliant example is concerned, I think you are right. I was trying to show that by continuously "resetting" the generator's instance you do not let the algorithm to work out. By the randomness perspective though, due to the use of CPU's time, it could be a good tactic to get better "real" random numbers, provided that an attacker cannot use/change system's time.

        • Your 1st compliant solution reseeds the RNG on each loop invocation. Which we agreed is 'random enough', but it doesn't match your discussion above the code.
        • Your last compliant solution works for me but only if 'number' is not static. Static members can not be declared within a method.
        1. I feel the first CS might be better as an exception. I always fear that someone might copy paste the CS code without reading the text properly.

          1. Our main criteria for making an exception to a rule is that the exception must be automatically. enforceable. IOW a sufficient rule checker should be able to tell when some code falls into the exception category. This can prob be done with annotations, or Javadoc comments.

            So the first CS can be moved into an exception if you think that is best.

            BTW the first CS would not fly in C or C++ because they don't specify how rand() works. Consequently, rand() need not be truly random (and there have been poor implementations in the past that cause security experts to not trust rand()). So seeding the standard RNG is only a compliant solution in Java.

            1. I've made the change. Thanks for the c/c++ ideas; I believe methods of class Random in Java were not meant for secure use even if they could be seeded (as described in the docs). So c/c++ & Java are alike in that respect.

              Incidentally, subclasses of Random are permitted to use their own algorithm provided they follow the general contract. (smile)

  1. Don't know how feasible it is to recommend generating strong random numbers outright. If you generate random numbers per client on a linux machine, the calls may block if the pool is exhausted (depends on /dev/random for entropy). Might be worth allowing precalculated seed data to be supplied from other pools/systems.

  2. SecureRandom on some operating systems can get very slow if their entropy source is depleted.  I've seen issues on RHEL 6 where our unit tests were taking 82 minutes compared to 5 minutes on Windows & other linux variants.

  3. Why is the Seveirty High? It strikes me that violation of this rule might compromise an encryption system based on an insufficiently random number. But that results in an information leak, not remote code execution. So the security should be Medium.

    1. Being able to predict a random number sequence may enable authentication to be bypassed, which could result in a system being compromised completely, including remote code injection.  The "Related Vulnerabilities" section refers to just such an attack.  Therefore, I think a severity of "High" is justified.

      1. OK. The analogous C rule is 'Medium', because we didn't consider privilege escalation when building the C standard.

  4. new SecureRandom() should be preferred over explicitly defining an algorithm as this rule does. Also verify this with the similar recommendation that does use new SecureRandom.

    The name of this rule should be revised or it should be mentioned explicitly that this isn't about getInstanceStrong as introduced in Java 8.

    As for depletion of the entropy pool, see which is integrated into Java 8.

    1. I've updated the rule as suggested. There is a new compliant example that uses the getInstanceStrong() method.