Hash Generator

A hash function is expected to hold three properties. Pre-image resistance: given a digest, finding any input that produces it is infeasible. Second pre-image resistance: given one input, finding a different input with the same digest is infeasible. Collision resistance: finding any two inputs that share a digest is infeasible. Collision resistance is the weakest of the three and the first to fall. It is what MD5 and SHA-1 have lost, while their pre-image resistance is still intact.

A hash is not encryption. There is no key and no decryption step. The mapping is deliberately lossy, so the original input cannot be recovered from the digest by any means other than guessing.

The answer depends entirely on whether an attacker controls the input, and on whether you are hashing data or a password. General-purpose hashes and password hashes are not interchangeable and choosing the wrong category is a more serious mistake than choosing the wrong algorithm within one.

No, not for anything security-relevant. MD5 collisions were demonstrated by Wang and Yu in 2004, and in 2008 a research team used a chosen-prefix collision to forge a certificate that chained to a real trusted CA. Chosen-prefix collisions against MD5 now cost seconds on ordinary hardware. The Flame malware weaponised an MD5 collision in 2012 to sign itself as Microsoft code.

MD5 must never be used for password storage, digital signatures, certificate verification, or any integrity check where an attacker could influence the input. It remains acceptable for exactly one category: detecting accidental corruption in a non-adversarial setting, such as verifying that a file copied across a network arrived intact. SHA-1 is in the same position and should be treated identically. Google's SHAttered attack broke its collision resistance in 2017, and chosen-prefix collisions followed in 2020.

Note what is not broken: no pre-image attack on either function is practical, so an MD5 or SHA-1 digest still does not reveal its input. The failure is collision resistance, which is what signatures and adversarial integrity checks depend on.

bcrypt is a password-hashing function derived from the Blowfish cipher, published by Niels Provos and David Mazières in 1999. It builds in a 16-byte salt, so the same password produces a different hash every time and precomputed rainbow tables are useless, and it takes a cost factor that sets how many iterations of key setup it performs. Each increment of the cost factor doubles the work.

OWASP's 2026 guidance is a cost factor of 12 as a minimum, with 13 or 14 preferred where the login latency budget allows. Tune it so a single hash takes roughly 250 to 500 milliseconds on your production hardware, and raise it again as hardware improves. That is what the tunable cost is for.

Two bcrypt limitations matter in practice. It silently truncates input at 72 bytes, so anything a long passphrase contains past that point counts for nothing. The usual workaround is to pre-hash with SHA-256 before bcrypt, but base64-encode the digest first or the NUL bytes in it will truncate the input further. bcrypt is also not memory-hard, which leaves it weaker than Argon2id against GPU and ASIC attack, and that is why Argon2id is the first choice for new systems.

Compute the hash of the file you downloaded and compare it character by character with the value the publisher lists. If they differ, the file is not the one that was published.

A hash function operates on bytes, not on characters, so the encoding of the input decides the output. The string café hashes differently as UTF-8 than as UTF-16 or Latin-1, and a Unicode string in NFC normalisation hashes differently from the same string in NFD. This generator hashes the UTF-8 bytes of what you type.

The other frequent causes of a mismatch are a trailing newline that a text editor or echo added (use printf or echo -n instead), and comparing a hex digest against a base64 digest of the same bytes.