Login
You're viewing the functional.cafe public feed.
  • Oct 16, 2026, 7:20 PM

    Anthropic found the origin of the NIST elliptic curve seeds: they were generated with drand48()!

    No one was seriously worried, but this should put some FUD to rest, completing the NIST ECC rehabilitation arc… just in time to be retired in favor of PQ.

    Also, we raised $36 Ki for ISRG. Yay!

    words.filippo.io/seeds-solved/

    💬 4🔄 29⭐ 7

Replies

  • Oct 16, 2026, 8:48 PM

    @filippo what a fun piece of computational archaeology (earnestly) and i hope the process doesn't contribute to flooding my home with seawater a few years earlier than it otherwise will be (also earnestly)

    💬 0🔄 0⭐ 0
  • Oct 16, 2026, 10:40 PM
    Those are not the curves I was looking for.

    I would not say the arc has been completed before we know where the Dual EC DRBG parameters came from. That algorithm may have been withdrawn, but every future cryptographic standard will still be seen in the light of that story.
    💬 1🔄 0⭐ 0
  • Oct 17, 2026, 7:25 AM

    @kasperd @filippo thats one interesting question, the other direction to continue this would be:
    Can LLMs help us learn how the AES SBoxes have been constructed?
    As far as I understand, they also have been provided by NSA without explanation.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Oct 17, 2026, 7:15 AM

    @filippo

    Anyone who attempts to generate random numbers by deterministic means is, of course, living in a state of sin.
    —John von Neumann (1951)

    💬 0🔄 0⭐ 0
  • Oct 17, 2026, 8:13 AM

    @filippo

    which I would now personally recommend over Curve25519/Ed25519/X25519.

    I'm very curious why you would recommend over the *25519. I understand the confidence to recommend the curves, but why over the other ones ?

    💬 1🔄 0⭐ 0
  • Oct 17, 2026, 11:40 AM

    @S1m @filippo

    1. Hardware implementations of P-256 remain far more widespread than Curve25519, and this is unlikely to improve particularly quickly (e.g. the Mac secure enclave doesn’t do 25519). Rather than deploying hardware for 25519, we should focus on deploying hardware for ML-DSA, ML-KEM & co.
    2. Lots of applications of EC require a prime-order group; the NIST P- series curves all provide this. Yes, you can use ristoretto255 to provide this over Curve25519, but its additional complication. If you’re not using Ristoretto, there are a few traps to watch out for in 25519 that aren’t there in P-256
    3. Most of of the improvements that e.g. EdDSA provides over ECDSA have been backported to modern implementations of the latter (e.g. modern ECDSA signers mix the to-be-signed value with their random number generator input to defend against randomness issues)
    💬 1🔄 1⭐ 0
  • Oct 17, 2026, 4:58 PM

    @erincandescent
    Thank you for the answer! The 2 first points are valid to me to explain why it could be recommended over the other one, but it's very context dependant

    @filippo

    💬 0🔄 0⭐ 0