diff --git a/crates/vrf_fun/RUSTSEC-0000-0000.md b/crates/vrf_fun/RUSTSEC-0000-0000.md
new file mode 100644
index 000000000..d22f131e7
--- /dev/null
+++ b/crates/vrf_fun/RUSTSEC-0000-0000.md
@@ -0,0 +1,43 @@
+```toml
+[advisory]
+id = "RUSTSEC-0000-0000"
+package = "vrf_fun"
+date = "2026-02-27"
+url = "https://github.com/LLFourn/secp256kfun/pull/244"
+categories = ["crypto-failure"]
+keywords = ["nonce-reuse", "key-recovery", "vrf"]
+license = "CC-BY-4.0"
+
+[affected.functions]
+"vrf_fun::Vrf::prove" = ["= 0.12.0"]
+"vrf_fun::rfc9381::tai::prove" = ["= 0.12.0"]
+"vrf_fun::rfc9381::sswu::prove" = ["= 0.12.0"]
+"vrf_fun::rfc9381::Rfc9381Transcript::gen_rng" = ["= 0.12.0"]
+
+[versions]
+patched = []
+```
+
+# RFC 9381 VRF nonce reuse allows secret key recovery
+
+The `ProverTranscript` implementation for `Rfc9381Transcript` computed the
+nonce seed from a fresh hasher rather than the transcript state. The nonce
+depended only on a fixed domain separator and the secret key, not on the VRF
+input:
+
+```
+nonce = H("vrf-nonce-gen" || secret_key)
+```
+
+Because the nonce was identical across all proofs made with the same key,
+any two VRF proofs on different inputs leaked the secret key via:
+
+```
+x = (s1 - s2) / (c1 - c2)
+```
+
+where `(c1, s1)` and `(c2, s2)` are the challenge-response pairs from the
+two proofs.
+
+The fix incorporates the transcript state (which includes the public key,
+hash-to-curve point, and gamma) into the nonce derivation.