File Hashes
imphash, ssdeep, TLSH, and so forth: what survives a change, and what a match lets you claim.
Every scheme here gets quoted as if it identified a malware family. Most identify something narrower, and a couple identify something else. Each row says what the hash is computed over, because that predicts everything else about it.
Before you read the table
- Exact hashes change if one byte changes. A match is not a judgment call.
- Similarity hashes are built to survive change, which is why a match is weaker evidence than it looks.
8 rows.
| Name | What a match means |
|---|---|
| MD5 | These are byte-identical files, unless someone constructed a collision. Still the most widely exchanged identifier in threat intelligence, which is why it persists despite being cryptographically broken. |
| SHA-256 | These are the same file, full stop. The right default for recording what you analyzed, and the one to quote when the claim needs to hold up. |
| Authentihash | The same binary content, regardless of who signed it or whether it was signed at all. Useful when an operator re-signs the same payload with a fresh stolen certificate: the file hash changes and this does not. |
| imphash | These binaries were built from a similar codebase with the same toolchain, OR they were packed with the same packer. Distinguishing those two cases is the entire skill. A shared imphash across unpacked samples is a genuinely strong family signal; across packed ones it is close to meaningless. |
| telfhash | The ELF equivalent of an imphash match. Built to fill the gap where imphash has no counterpart, and it inherits TLSH's distance score rather than being a plain equality check. |
| ssdeep (CTPH) | A percentage similarity, not a yes or no. A high score means large runs of identical bytes. It does NOT mean similar behavior, and two samples of one family compiled from the same source with different settings often score zero. |
| TLSH | A distance, where LOWER is more similar and zero is identical: the opposite direction from ssdeep's percentage, which is a constant source of misread numbers. Generally more reliable than ssdeep for clustering, and the scheme telfhash is built on. |
| Rich header hash | These binaries were built on similarly configured machines with the same toolchain versions. That is an attribution-flavored signal about the DEVELOPER rather than a similarity signal about the code, which makes it powerful and easy to overstate. |