Skip to content

The PQ situation at Tor

  • discussion started in 2016 with Isis's/Peter Schwabe's proposal

    • NewHope handshake
    • lost steam within Tor
  • ML-KEM got specified by NIST

  • later in Tor proposal 355

    • looking at Tor in mutliple layers
      1. TLS layer
      2. circuit layer
      3. signarures/onion services (active attacker)
    • we are not concerned with the third case at this point as the PQ schemes here have very large keys and signatures and an attack here is unlikely right now
  • openssl 3.5 has ML-KEM support in hybrid mode

  • situation at Tor relays

    • in February 2026 31.2% of relays supported PQ handshakes
    • right now (November 2026) the number has grown to 62.6%
  • libressl got support as well; we need to figure out, though, whether there is some cipher suite PQ overlap in the openssl/libressl space

  • circuit layer PQ work depends on pack/frag proposal in arti where we can split up 509 byte cells

    • we can control the KEM part here, but need to use whatever is available in the TLS space, so we could theoretically use something different for the circuit layer
    • do we want to use a pure one instead of the hybrid one?
  • for signatures the situation is very open; for onion services there is no design either available yet
  • there is no timeline for the circuit layer/signatures yet, only TLS is considered now
  • next step would be to apply for funding to fix the circuit layer
  • focus only on openssl/libressl; boringssl doesn't do any releases