Ethereum research asks which hashes people have actually attacked
Serguey Asael Shinder1 min readA post by khovratovich on ethresear.ch, dated 23 September 2026, works through which hash functions Ethereum should now prefer and why the answer has changed: https://ethresear.ch/t/post-poseidon-hash-function-variants-for-ethereum/26071.
The premise is the change. With Flock available as a post-quantum proof system for binary circuits, the protocol no longer needs hashes chosen for being circuit-friendly, which is the property the Poseidon family exists to supply.
The comparison that follows is about cryptanalytic maturity rather than speed. BLAKE2 has no published collision or preimage attack at any round count, but carries a short record - fewer than ten cryptanalysis papers - which the post attributes partly to design strength and partly to lower attention. Poseidon variants and Tip5 remain unbroken while analysis against them is still advancing quickly, and the post declines to call them mature. RIPEMD-160 is noted as a second layer in Bitcoin address derivation whose short output makes it unsuitable elsewhere.
What it means
A proof system changing can retire a whole class of design constraint, and that is the news here. Circuit-friendly hashes were an answer to a cost: proving a SHA-2 computation used to be prohibitive. If that cost falls, the case for accepting a younger construction with a thinner literature falls with it.
"Unbroken" and "mature" are different claims, and the post keeps them apart. No attack published means nobody has succeeded; a short cryptanalytic record means few have tried hard enough for long enough. For a protocol with no upgrade path once the hash is embedded in signatures and state, the second question is the one that decides.
⚠️ This is a research forum post, not a protocol decision. Nothing here is scheduled, and the summary table the author provides is an assessment offered for discussion.