MD5 Is Hashing, Not Encryption: How It Works, How It Gets Cracked, and How to Store Passwords Correctly

What MD5 Actually Is

Many people call MD5 "MD5 encryption." That framing is wrong from the start.

MD5 is a hash function, not an encryption algorithm. The distinction is fundamental:

• Encryption: A reversible operation with a key — the original data can be recovered by decrypting
• Hashing: A one-way operation with no key — the original data cannot theoretically be recovered from the output

MD5 takes an input of any length and produces a fixed 128-bit (32 hexadecimal character) output called a "digest" or "fingerprint."

MD5("hello")         = 5d41402abc4b2a76b9719d911017c592
MD5("hello!")        = 9b56e4f280d7b27a6c8d0c11d5a8cb7d
MD5("a very long...") = d41d8cd98f00b204e9800998ecf8427e

Three core properties:

1. One-way: The original input cannot be derived from the hash value
2. Fixed length: Output is always 32 characters regardless of input size
3. Avalanche effect: Changing a single byte in the input produces a completely different hash

Can MD5 Be "Cracked"

Strictly speaking, MD5 cannot be "decrypted" — because it is not encryption. But it can be broken through brute force or lookup table attacks.

Brute Force

The simplest approach: compute the MD5 of common passwords one by one and compare against the target hash.

Compute MD5("123456") = e10adc3949ba59abbe56e057f20f883e  ← match!

A modern GPU can compute billions of MD5 hashes per second. Common passwords like 123456, password, and qwerty are found in under a second.

Rainbow Table Attacks

Brute force requires real-time computation. A rainbow table is a precomputed database of plaintext-to-hash mappings — trading storage space for speed.

An attacker with a target MD5 hash simply looks it up in the table and instantly retrieves the original password. Public rainbow tables covering tens of billions of common passwords and short strings are freely available online.

This is exactly how most "MD5 online decryption" websites work — they maintain a massive hash-to-plaintext database and return results via lookup, not true decryption.

MD5 Collision Attacks

In 2004, researchers demonstrated that MD5 collisions could be deliberately constructed: two completely different inputs that produce an identical MD5 hash.

File A and File B have entirely different contents, yet produce the same MD5 hash

This means an attacker can craft a malicious file with the same MD5 as a legitimate file, bypassing integrity checks entirely. This fundamentally undermined MD5's trustworthiness even for non-password use cases.

Why Systems Still Use MD5 for Passwords

History. MD5 was published in 1991 and was considered secure at the time, leading many early systems to adopt it for password storage. As computing power grew and attack methods evolved, MD5's security became wholly inadequate — but legacy system migration is expensive, so MD5 persists in many codebases today.

In the 2012 LinkedIn breach, roughly 6.5 million passwords were stored as unsalted SHA-1 hashes. Earlier breaches exposed systems using bare MD5. In both cases, attackers cracked the overwhelming majority of passwords within hours of obtaining the database.

The Right Way to Store Passwords

Salting

Append a random string (the salt) to each password before hashing, so that identical passwords produce different hash values — rendering rainbow tables useless.

salt = "xK92mP"  // randomly generated per user, stored in the database
hash = MD5(password + salt)

However, even with salting, MD5's raw computation speed still makes GPU brute-force attacks viable.

bcrypt: Purpose-Built for Password Storage

bcrypt is the most widely adopted password hashing scheme today, with several critical design properties:

1. Built-in salting: A random salt is generated automatically on every hash — no manual handling needed.

2. Adjustable cost factor: A rounds parameter controls how computationally expensive each hash operation is, and can be increased over time as hardware improves.

3. Deliberately slow: A legitimate user verifying their password only needs one hash computation — the slowness is negligible. An attacker attempting billions of guesses finds the slowness devastating.

import bcrypt

# Registration: generate hash
password = b"user_password"
hashed = bcrypt.hashpw(password, bcrypt.gensalt(rounds=12))
# Store hashed in the database — never the raw password

# Login: verify
bcrypt.checkpw(password, hashed)  # returns True or False

Argon2: The Modern Standard

Argon2 won the Password Hashing Competition in 2015. It extends bcrypt's design with configurable memory usage, making GPU and ASIC brute-force attacks significantly more expensive. New projects should default to Argon2id.

from argon2 import PasswordHasher

ph = PasswordHasher(time_cost=2, memory_cost=65536, parallelism=2)
hashed = ph.hash("user_password")
ph.verify(hashed, "user_password")  # verification

scrypt

Similar to Argon2 in its memory-hard design. Built into the Node.js standard library, making it a natural fit for JavaScript backends.

Where MD5 Is Still Acceptable

MD5 is unsuitable for security-sensitive scenarios, but remains reasonable in the following non-security contexts:

File integrity checking (non-adversarial)

Comparing MD5 values after downloading a file to confirm it was not corrupted in transit. Note: this does not protect against intentional tampering — only accidental corruption.

# Linux / macOS
md5sum filename.zip

# macOS alternative
md5 filename.zip

Generating unique identifiers

Using MD5 of content as a cache key, deduplication ID, or similar identifier where security is not a concern.

Database content deduplication

Hashing large text fields with MD5 for indexed deduplication lookups.

Non-security data fingerprinting

Quick content summaries in logging systems and data pipelines for tracking and comparison.

Summary Table

Never use MD5 for password storage — with or without salting.

Online Tool

To compute the MD5 hash of a text string for data verification or testing purposes, the MD5 Online Encryption Tool on toolshu.com supports both 16-character and 32-character output in uppercase or lowercase. All computations run locally in your browser with no data uploaded.