RIPEMD-160 collision at a 48-bit prefix

Result and evidence

The two distinct eight-byte inputs in collision.json produce these RIPEMD-160 digests:

Input	Bytes (hex)	Full RIPEMD-160 digest (hex)	First 48 bits
A	19bef10000000000	902e5fc86be47b1e2a17de2aab65d4e765623a05	902e5fc86be4
B	1022730100000000	902e5fc86be4a99253faddc650510cb85e77233d	902e5fc86be4

Observed facts: The byte strings differ, and both hashlib.new('ripemd160', ...) in Python 3 and openssl dgst -ripemd160 -binary using OpenSSL 3.0.13 returned the full digests shown above. OpenSSL documents its RIPEMD-160 digest implementation; its low-level RIPEMD160 function computes a 20-byte digest.

Inference: The first six digest bytes are identical, so truncating either digest to its first 48 most significant bits yields the same value, 902e5fc86be4. The full 160-bit digests differ; this is a collision only for the specified truncation.

Method: A local C search hashed successive unsigned 64-bit integers encoded as eight bytes in little-endian order. It indexed the first six digest bytes and stopped on the first repeated value, after 24,322,577 inputs. The search used OpenSSL's RIPEMD160 function; the verification above used its command-line digest interface and Python's hash interface. The result can be reproduced without the search:

import hashlib

a = bytes.fromhex('19bef10000000000')
b = bytes.fromhex('1022730100000000')
da = hashlib.new('ripemd160', a).digest()
db = hashlib.new('ripemd160', b).digest()
assert a != b
assert da[:6] == db[:6] == bytes.fromhex('902e5fc86be4')
print(da.hex(), db.hex())

Uncertainty and unanswered questions: None affect the stated collision: the evidence is the reproducible digest calculation on the exact submitted bytes. This report makes no claim of a collision in the full RIPEMD-160 output.