IETF LLM rules, ML-KEM TLS last call, prover bugs

IETF lists saw a last call dispute over pure ML-KEM for TLS 1.3, a proposal on LLM text in standards work, and adoption calls plus modeling fights in adjacent groups. Formal verification bugs and OpenPGP hardware details rounded out the day.

LLM-generated text in IETF discussions

IETF participants debated a draft on handling LLM-generated text in standards discussions. The exchange focused on accountability and disclosure. Developers who contribute to or rely on IETF work need clear rules as LLM use spreads in protocol design.

ML-KEM post-quantum key agreement last call

The IETF opened last call on draft-ietf-tls-mlkem-09 for ML-KEM post-quantum key agreement in TLS 1.3 as an Informational RFC. Bernstein-led opposition cited 82 objectors, consensus failures, and preference for hybrid security. TLS implementers and crypto library maintainers should track whether pure ML-KEM advances or stalls.

Soundness bugs in Tamarin prover

Nadim Kobeissi reported critical soundness bugs in the Tamarin prover discovered via LLM, following earlier Lean prover flaw discussion on the UFMRG list. The findings affect tools used to verify security protocols. Protocol designers who depend on automated provers have fresh reason to recheck their models.

OpenPGP cards and v6 key packets

OpenPGP list participants discussed an encoding scheme for 20-byte card fingerprint fields so existing hardware security devices can support v6 OpenPGP keys. The thread examined compatibility with modern key packets. Hardware token users and OpenPGP implementers need a workable path for newer key formats on current cards.

AI agent authentication adoption call

The WIMSE working group issued a call for adoption of a draft on AI agent authentication and authorization. Mixed votes and technical debate covered scope, agent versus workload identity, and LLM mission handling. Readers building or securing AI systems should note how IETF identity work is expanding into agent scenarios.

Early attestation draft and CVE claim

Multiple SEAT list participants asserted that CVE-2026-33697 still applies to draft-fossati-seat-early-attestation-06, citing ProVerif models that show a failed G3 property. Authors disputed the threat model scope. Attestation protocol implementers must weigh the modeling results against the authors' narrower framing.

OAuth RAR metadata adoption call

The OAuth working group called for adoption of the OAuth 2.0 RAR Metadata and Error Remediation draft. Fourteen participants gave unanimous early support. OAuth developers gain clearer metadata and error handling for rich authorization requests if the draft proceeds.

CDDL validation and bignum tags

CBOR list participants disagreed on whether CDDL #0 matches bignum tags under RFC 8949 unification, exposing validator divergence for RFC 8610. The dispute turns on how number forms are treated. CBOR and CDDL tool authors need consistent rules to avoid silent interoperability failures.