MD5 vs SHA-256: What Hashes Prove and Which to Use

#What a hash actually proves

A hash function compresses data of any length into a fixed-length digest, and the process cannot be reversed. Three properties follow from that:

1. Deterministic - the same data always produces the same digest. Change one byte and the entire output changes.
2. One-way - you can compute a digest from the input, never the input from the digest. So-called "hash cracking" is dictionary lookup: hash candidates and compare. It is brute force, not mathematics.
3. Collision-resistant - it should be infeasible to find two different inputs with the same digest. How strong that guarantee is separates MD5 from SHA-256.

That is why a checksum answers exactly one question: *is this the same data I expected?* Did the download complete, did the transfer corrupt anything.

#Digests and their lengths

Length is your quickest sanity check - if a value is 32 characters, no SHA-256 produced it.

#Verify a download in three steps

1. Open SHA Hash, or MD5 Hash if the publisher only provides MD5.
2. Drop the file into the upload area. It is read and hashed locally in your browser, so the file is never uploaded.
3. Compare against the published checksum character by character. Partial matches mean nothing.

On the command line you should get an identical result:

# macOS / Linux
shasum -a 256 package.tar.gz
md5sum package.tar.gz

# Windows PowerShell
Get-FileHash package.tar.gz -Algorithm SHA256

#Why salting is not optional for passwords

Unsalted hashes fail for passwords because the input space is tiny, not because the algorithm is weak. Common passwords number in the hundreds of thousands, so an attacker can precompute a plaintext-to-digest table (a rainbow table) and look up any hash in milliseconds.

Salting prepends a unique random string per user before hashing, so identical passwords produce unrelated digests across accounts and sites:

#A UUID is not a hash

These two appear together often and are unrelated:

• A hash is a digest of existing data. Same input, same output, fixed length.
• A UUID is a generated identifier with no input. Use UUID / GUID Generator to assign unique IDs to database rows, sessions or requests.

Need a unique identifier? Use a UUID. Need to detect whether data changed? Use a hash. Using a hash as an ID brings two problems: collisions are theoretically possible, and if the ID is derived from user input it becomes predictable.

#Cheat sheet

#Frequently asked questions

#Can a hash be decrypted?

No, and "hash decryption" does not exist as an operation. Sites claiming to decrypt MD5 work by storing a huge lookup table of plaintext-to-digest pairs. Your hash is shown as plaintext only if it is already in the table. Short, common passwords "decrypt" because they are in the dictionary, not because the algorithm was broken.

#Is MD5 still usable at all?

It depends on the threat model. For integrity against accidental corruption - truncated downloads, damaged transfers - it works, because those failures are random. For anything adversarial - signatures, certificates, tamper protection, password storage - it does not, because collision attacks are public and practical.

#Why do I get a different hash for the same file on Windows and Linux?

Suspect the file first. Text-mode transfers on Windows rewrite \n to \r\n, which changes the bytes and therefore the digest. Re-transfer in binary mode. More rarely, two commands default to different algorithms - pass the algorithm explicitly to be sure.

#Is SHA-256 much slower than MD5?

Not in any way that matters for everyday use. On modern CPUs SHA-256 is roughly twice as slow, which on a tens-of-megabytes file is still a matter of milliseconds to a few hundred. Choose on security grounds, not speed.

#Does a longer hash always mean better security?

Up to a point. SHA-512 has a longer digest and a higher theoretical strength, at the cost of longer values to store and transmit. SHA-256 is the practical default, and going longer is rarely necessary outside specific compliance requirements.