What is the difference between a "normal" random number generator and a cryptographically secure one (CSPRNG)?

Question

Accepted Answer

A normal RNG produces uniformly distributed numbers; a CSPRNG additionally guarantees its output is unpredictable — even given all previous outputs, the attacker can't guess the next one.

Property
Normal RNG
CSPRNG

Uniform distribution
✅
✅

High throughput
Usually ✅
Often slower

Output unpredictable to attacker
❌
✅

Suitable for simulations, games, random sampling
✅
✅

Suitable for keys, IVs, tokens, salts
❌ NEVER
✅

Java examples:

Use case
Class

Normal
java.util.Random (linear congruential — predictable from 1–2 outputs)

Cryptographic
java.security.SecureRandom (CSPRNG, OS-seeded)

Other CSPRNGs by platform:

Platform
Cryptographic source

Linux
/dev/urandom (or getrandom(2))

Windows
BCryptGenRandom / CNG

macOS/iOS
SecRandomCopyBytes / CCRandomGenerateBytes

Python
secrets module

Node.js
crypto.randomBytes (never Math.random())

Go
crypto/rand (never math/rand)

Tip: If a code review shows Math.random(), rand(), time(NULL), or mt19937 being used to generate keys, IVs, session IDs, password-reset tokens, etc. — that's a real finding. The output is recoverable from a few observations and the attacker can predict future tokens.

Property	Normal RNG	CSPRNG
Uniform distribution	✅	✅
High throughput	Usually ✅	Often slower
Output unpredictable to attacker	❌	✅
Suitable for simulations, games, random sampling	✅	✅
Suitable for keys, IVs, tokens, salts	❌ NEVER	✅

Use case	Class
Normal	`java.util.Random` (linear congruential — predictable from 1–2 outputs)
Cryptographic	`java.security.SecureRandom` (CSPRNG, OS-seeded)

Platform	Cryptographic source
Linux	`/dev/urandom` (or `getrandom(2)`)
Windows	`BCryptGenRandom` / CNG
macOS/iOS	`SecRandomCopyBytes` / `CCRandomGenerateBytes`
Python	`secrets` module
Node.js	`crypto.randomBytes` (never `Math.random()`)
Go	`crypto/rand` (never `math/rand`)