True Randomness vs Pseudorandomness

Cryptographic security depends fundamentally on unpredictability. If an attacker can predict your key, IV, or nonce — even partially — the entire cryptographic system may collapse. Weak randomness has caused catastrophic real-world breaks.

TRNGs harvest physical entropy: thermal noise in resistors, shot noise in reverse-biased diodes, radioactive decay, mouse movements, keystroke timings, disk seek latencies. These are genuinely unpredictable physical phenomena.

The OS maintains an entropy pool — a collection of physical noise samples XORed and hashed together. Linux: /dev/random and /dev/urandom draw from this pool. The pool is replenished continuously as the system runs.

A PRNG (also called a Deterministic RNG / DRNG) expands a short seed into a long stream of pseudo-random bytes using a deterministic algorithm. Given the same seed, it produces the same output. Not suitable for cryptography unless seeded with true entropy.

A PRNG may have detectable statistical bias or be predictable from observed output. A Cryptographically Secure PRNG (CSPRNG) adds: (1) forward security (past output not predictable from current state) and (2) backtracking resistance (future output not predictable from past).

A PRNG is cryptographically secure if no polynomial-time algorithm can distinguish its output from a truly random sequence with probability significantly better than 1/2. This is the Next-Bit Unpredictability criterion.

Linux: /dev/urandom (non-blocking, after initial seeding), getrandom() syscall. Windows: CryptGenRandom(), BCryptGenRandom(). macOS/iOS: SecRandomCopyBytes(). Browser: window.crypto.getRandomValues(). All feed into the OS entropy pool.

NIST SP 800-22 provides 15 statistical tests for randomness: frequency, block frequency, runs, Maurer's universal, linear complexity, etc. Passing statistical tests does not prove cryptographic security — just absence of obvious bias.

A CSPRNG seeded with weak entropy (e.g., current time in seconds) is predictable. An attacker who knows the seed reconstructs the entire output. Always seed from OS entropy: os.urandom(32) in Python, not time.time() or random.random().

Intel RDRAND (x86 instruction) returns hardware-generated random bytes from a TRNG on-chip. AMD and ARM equivalent hardware exists. Linux mixes RDRAND into /dev/urandom. Used as supplemental entropy, not sole source (due to trust concerns).

QRNGs use quantum-mechanical phenomena (photon arrival times, vacuum fluctuations) for provably random output. Commercial QRNGs (ID Quantique, Quantis) are used in high-security applications. Cost prohibitive for general use.

What property distinguishes a CSPRNG from a regular PRNG?

TRNGs harvest physical entropy; PRNGs expand a seed deterministically; CSPRNGs are indistinguishable from true randomness. OS entropy pools combine both. Next: specific CSPRNG designs used in practice.