ML-KEM last call and IETF AI process debates

IETF traffic on 2026-08-13 focused on post-quantum TLS choices and on how large language models intersect with standards work. Heated objections to the ML-KEM draft sat alongside technical debates on LLM disclosure, theorem-prover soundness, and AI agent identity.

ML-KEM TLS draft enters last call with public objections

The IETF opened last call on draft-ietf-tls-mlkem-09, which defines ML-KEM post-quantum key agreement for TLS 1.3 and is aimed at Informational RFC status. D.J. Bernstein objected on cost, security, and process grounds. The result will shape how TLS deployments adopt quantum-resistant key exchange.

Draft on handling LLMs in IETF discussions

Participants examined a proposal for managing LLM-generated text in standards discussions, centering on accountability and disclosure. The exchange ran to 64 messages from 24 contributors and stayed technical. Clear norms would affect attribution and reliance on machine-generated input in future IETF work.

Soundness bugs reported in Tamarin prover

Nadim Kobeissi reported soundness bugs in the Tamarin theorem prover that an LLM helped surface. Maintainers acknowledged fixes yet disputed the classification of the issues. Prover reliability underpins formal verification of security protocols, so the dispute bears on tool trust.

WIMSE call for adoption of AI agent authentication

The WIMSE working group issued a call for adoption of a draft on AI agent authentication and authorization. Votes were mixed, with debate on scope, agent versus workload identity, and LLM mission handling. Adoption would influence identity models for autonomous agents in IETF protocols.

Bernstein disputes NSA influence claims in SSH thread

D.J. Bernstein contested claims of NSA influence over IETF ML-KEM/ML-DSA and hybrid-mode recommendations in an SSH working-group thread that began as a complaint to the chairs. The short exchange continues scrutiny of how post-quantum algorithm choices reach SSH.

OpenPGP cards and modern key packets

Contributors discussed an encoding scheme for 20-byte card fingerprint fields so existing OpenPGP card hardware can support v6 keys. Eight participants offered technical proposals. The work seeks to keep hardware security devices usable as OpenPGP key formats evolve.

CDDL bignum matching and validator divergence

A CBOR thread disagreed on whether CDDL #0 matches bignum tags under the unification in RFC 8949, exposing differences among validators. The discussion remained technical across 28 messages. Consistent rules matter for interoperable CBOR schema checking.

Trimming controversial features from EDN literals draft

The CBOR working group weighed removal or repair of the CRI app-extension, the unknown-app CPA999 tag, and block comments from the EDN literals draft to reach consensus. Five participants took part. Narrowing the feature set could unblock progress on the document.