PQC-XXJTLU · cryptanalysis

HomeHash functions › Neulaser

Hash function · symmetric (hash)

Neulaser

Design

neulaser-1

The oldest register word is read once and discarded: collisions, second preimages, multicollisions

Practical · Demonstrated · Cause: construction

Property: collision and second-preimage resistance

Cost against claim: collision in ≈216.3 feedback evaluations (under a second), against 2256 to 2512

In every round, the oldest word of each register module is read only by the feedback function and is then shifted out. Two states that differ only in that word and give the same feedback value are equal one round later. The message fills whole modules, oldest word included, so a birthday search over a single 32-bit message word yields full-digest collisions from any chaining value.

Parameter setClaimAttack costVerdict
Neulaser-512, -768, -1024collision 2256 to 2512; second preimage ≈2512 to 21024collision ≈216.3 feedback evaluations; second preimage ≈217Practical

What causes it

The specification itself notes that a concrete bound must add the generator's distinguishing advantage; that advantage is close to 1 after 216 queries.

Evidence

Cause in the specification: the attack also works on a model built from the specification alone.

Limits

The two messages have equal length and differ in one 32-bit word that lands in an oldest register stage. A second preimage succeeds with probability ≈63% per usable word. Preimages are not affected. The specification never states the loading order of its inputs; 6 of the 8 conceivable orders are vulnerable, and the other 2, which contradict the notation and the test vectors, allow free-start collisions only.

What PQC-X adds

Second preimages of the official test-vector messages, found by meet-in-the-middle in about 217 thanks to the separability of the feedback function; collisions from any chaining value, not only from the fixed IV; and the check of every loading order the specification leaves open.

Credit

First public report: Cryptanalysts001 (ISCAS) (CryptHash forum, also ngcc.dev hash-21-1, 2026-09-22). Found independently by PQC-X.

Related public reports: