IETF Daily Digest — Internet-Drafts & RFCs

A YANG Data Model and RADIUS Extension for Policy-Based Network Access Control

Describes a YANG data model for controlling network access based on group membership, adding date and time settings to Access Control Lists (ACLs) so rules can take effect on a schedule. It is aimed at environments where network access begins when a user logs in, simplifying the mapping from user group to the packet-header details used to enforce policy. It also defines a RADIUS attribute that carries the user group identifier within identification and authorization information, so that group-based, time-aware access control is easier to manage.

Computing-Aware Traffic Steering (CATS) Problem Statement, Use Cases, and Requirements

States that spreading computing work across facilities can shorten response times and save energy for demanding, time-sensitive services, and introduces Computing-Aware Traffic Steering (CATS), which selects servers and steers traffic using each server's processing power and available resources. Because current methods rely on fixed assignments or connection measurements alone, CATS aims to improve both throughput and response time. The document sets out the problem CATS addresses with example scenarios, limited to a single network domain, and concludes with the requirements a CATS framework should meet.

A Framework for Computing-Aware Traffic Steering (CATS)

Defines a framework for Computing-Aware Traffic Steering (CATS), which steers traffic based on both computing resources and network conditions. It specifies a set of CATS functional components and describes how they interact, and includes illustrative workflows showing how the control plane and data plane operate. The framework's scope is limited to a single service provider.

Proposes expanding the IPv6 loopback space from a single address to a /96 prefix, giving far more loopback addresses for inter-process communication and network diagnostics. A newly allocated TBD/96 prefix would be used rather than carving from the existing ::/96. The change updates the IANA IPv6 Address registry, the IPv6 Special-Purpose Address registry, and RFC 4291.

Specifies the observable behaviour of the CATS Service Metric Agent (C-SMA), the component that collects service capability and status information and reports it to CATS Path Selectors. It fills gaps left by the CATS framework and metric documents: how reported metrics are validated and cleaned, how stale or failing Service Contact Instances are handled, and how the agent stays synchronized with the Path Selector. It covers collecting reports, validating them to keep erroneous or malicious data out of steering, soft-state caching with aging, local and update control policy, and state reconciliation. An internal component breakdown is offered only as illustrative guidance.

Defines learnlog, a file format for recording what learners do in educational software. Each record holds typed events describing a learner's actions and the responses of people or programs, with entries chained by cryptographic hashes and optionally signed by the producer. Anyone trusting the signing key can detect changes to signed entries, though the security section notes some tampering can still go undetected. Individual events can be redacted while the rest of the record still verifies. The document specifies the file/record/entry/event structure and registries for event schemas, but excludes subject-specific schemas.

Describes how to distribute performance-measurement data for Traffic Engineering (TE) Policies through BGP-Link State (BGP-LS). Controllers can use these measurements to place services and use network resources more efficiently. It reuses the delay, delay-variation, and loss metrics from RFC 8571 for TE Policies and adds round-trip versions of those metrics.

Defines the Sovereign Validation Protocol (SOVP), a method for checking a signed identity document before a system accepts its data. A consumer fetches a JSON document from a standard location on a host, canonicalizes it, and verifies the signature against an Ed25519 public key published in a DNS TXT record for that host. The spec covers which parts of the document are signed, how documents and keys are retrieved, freshness checks, and limits on unverified input, plus an optional extension binding a document to a specific deployment instance. It explicitly does not prove legal identity or guarantee content accuracy.

Extends STAMP, a two-way active performance-measurement protocol, so that the receiving (reflector) device returns the IPv4 or IPv6 headers, plus any IPv6 extension headers, from the test packets it received. This lets the sender see how packets were handled along the path. IOAM data fields are presented as one example of the information that could be carried, complementing STAMP's end-to-end measurement with per-hop and path telemetry.

Proposes changing the IETF Nominating Committee (NomCom) so its voting members are not all of one gender, building on the RFC 8713 rule that limits how many members share one organization. Volunteers who opt into a self-declared group could be guaranteed up to five voting seats, but only in years when the standard random selection would otherwise give that group fewer seats. In all other years the selection proceeds as usual.

Defines a YANG data model for managing network incidents end to end, where an incident is a problem that disrupts or degrades a network service. The model gives systems a common format for reporting incidents and supports diagnosing root causes. By grouping related issues into a single incident it aims to cut the number of troubleshooting tickets operators handle, resolve problems faster, and protect service health.

Predicts that scale-up, scale-out, and scale-across AI computing networks will converge into one shared fabric and proposes a way to compress Layer 3 packet headers in such a network. It assumes IPv6 as the Layer 3 protocol over a single shared fabric. For packets that remain within one super-node, the header can shrink to 8 octets, cutting header overhead by roughly 80%. The draft covers the motivation, requirements, benefits and feasibility, and the proposed header format.

Defines how the SQIsign post-quantum signature scheme is used with the COSE and JOSE formats common in constrained and IoT settings. SQIsign produces very small signatures and public keys, which the authors argue could help with limited authentication hardware such as FIDO2 CTAP2 devices. It warns SQIsign is still under evaluation in NIST standardization, so the underlying method may change, and notes SQIsign does not expose the torsion-point data behind the 2022 SIDH/SIKE attacks, though isogeny schemes remain under cryptanalysis. Stable identifiers are provided for interoperability.

Describes Key Transparency, a way to ensure that users' encryption keys in secure messaging services are genuine. It defines key terms, common deployment models for group messaging, and the security guarantees the protocol provides, plus non-binding guidance on applying the approach to common application types. The mechanism prevents a service provider from quietly substituting a user's key, because every change is recorded in a log users can audit.

Defines records that let anyone later verify, offline, what an automated agent was permitted to do and what it logged doing. A person approves a permission with a passkey, and the permission carries machine-checkable limits. The agent signs a receipt for each action using both a conventional and a quantum-resistant signature method, and receipts go into a log whose signed summary is time-stamped. A verifier reports whether each action stayed within the permission, kept separate from whether the records themselves are sound.

Notes that ad hoc networks struggle with IPv6 addressing because interface-to-neighbour connectivity is unpredictable, so nodes need addresses that are locally unique and independent of topology, especially when normal address-assignment services are absent or intermittent. Introduces a new IPv6 address type, the Multilink Local Address (MLA), which nodes can create and assign to their own interfaces without outside help. This supports ad hoc operation as devices join, leave, or split apart.

Aims to make payments by AI agents auditable, building on HTTP 402, AP2, and card-processing systems with a cryptographically secured reconciliation method. Before money moves, the payee issues a signed offer called a Trade Manifest, and a Policy Decision Point checks each spend request against stored rules, denying anything not explicitly allowed. After a gated payment, a signed Spend Receipt records what happened, and receipts are later checked against the payment rail's own settlement records.

Observes that online access today splits into a login that proves identity and a separate mechanism that protects the resulting token, leaving a gap because the login never establishes the session-protecting key. Proposes authenticating the user during the TLS handshake using a device-held key that can sign only while the user is present, the device is enrolled, and its state meets policy. Using that key as the TLS client key makes the handshake the login and the TLS session the session, with no separate login or session token; if a condition fails the key stops working and access ends at the next full handshake, without a revocation message. The authors stress this is not an OAuth extension.

A revision of the core BGP-4 specification that replaces RFC 4271. BGP connects Autonomous Systems and shares reachability information, including the AS path, which is enough to build network-level connectivity maps, remove routing loops, and apply policy. It describes BGP-4 support for CIDR, letting routers advertise address prefixes and dropping the old class-based address scheme, and the ability to aggregate multiple routes (and their AS paths) into summary routes.

By its name, this IETF PLANTS working-group draft concerns 'Merkle Tree Certificates,' a certificate format that most likely uses Merkle trees to provide compact, efficient authentication well-suited to a post-quantum setting, probably by batching issuance so that proofs and chains are smaller. The official abstract could not be retrieved on this run (the fast page returned 404), so this description is approximate and will be grounded on the next run.

Describes a simple, extensible routing system for the future global aviation network (ATN/IPS) that ICAO is studying, which would use IP to supplement existing communications for air traffic controllers, airline operations, and commercial aircraft worldwide. The design handles routing for mobile aircraft and is built on two established internet technologies: BGP for routing and DNS for lookups. It is presented as a way to meet the network's requirements.

Defines the 'moqt' URI scheme that identifies Media over QUIC Transport servers and explains how such URIs are resolved and how clients discover servers. It covers the URI syntax, fragment identifiers, how a client uses a URI to connect, and URI normalization rules, plus how a server's X.509 certificate is checked against a moqt URI. Server discovery uses DNS SVCB and SRV records as well as local-network mDNS and DNS-SD.

Proposes an experimental set of handling rules for email receiving systems that use a companion AI-processing preference signal. Before any AI can access an email, the system must check the sender's stated preference, with a separate check for each type of use, and when no preference is stated it protects content by default rather than assuming permission. It covers confirmation, messages that mix content from several sources, and records derived from emails, while keeping ordinary delivery usable for anyone declining AI processing. The rules apply only to systems claiming the profile and do not determine legal consent.

Defines an experimental email header that lets senders state how AI tools may use their messages, with categories such as AI assistance, long-term AI memory, profiling of people, model training, and forwarding to external processing services. It specifies the header structure, what each category means, which parts of a message it covers, and how it is preserved on forward or reply. A missing header is treated as no stated preference, not permission; recipient access controls and the default-protective behaviour live in a companion document. The header does not create legal consent or guarantee recipients will honour it.

Addresses the fact that once release files and their cryptographic fingerprints are fixed, attribution can't be changed by editing them without disturbing the evidence. Proposes a separate, versioned companion record that links two unchanged source fingerprints to named attribution and an opaque reference to a private origin record, using standardized JSON serialization, a fixed message prefix, and Ed25519 signatures. Verification checks five things independently: data integrity, signer authorization, source match, revision currency, and private-reference resolution. The authors frame it as a discipline for preserving and reporting evidence, not a new cryptographic method.

Changes the AUTH48 (or equivalent) step of RFC publication by adding rules that the IETF Datatracker enforces automatically, including built-in validation checkpoints and access controls meant to stop major technical edits after Working Group Last Call. The stated purpose is to protect the group's rough consensus, and the automated rules are meant to apply across all current and future publication streams. The document updates RFC 6359 and RFC 7841.

Argues that a controller managing edge-cloud services needs measurements whose meaning holds up across different collectors, pipelines, and administrative domains, because a metric's name and value alone don't show whether it can be compared, combined, or trusted at a given moment. Proposes a compact profile tying each KPI definition to what it observes, the population it covers, its time window, its freshness, and the evidence for its quality. It gives calculation rules for five KPI families, a procedure for deciding whether a consumer may use an observation, and worked examples, complementing existing telemetry work rather than defining a transport, YANG module, or IANA registration.

A formal specification of the Domain Connect Protocol, which lets service providers (hosting, email, social platforms, hardware) configure DNS records on a user's domain by passing configurations between the service company and the DNS provider. It targets the common problem that manual DNS setup is error-prone and confusing, causing support load and abandoned setups, and automates it through user-friendly web interactions, standard authentication, and templates.

Proposes a two-layer reference architecture, developed in the IETF DAWN working group, for discovering agents across organizational boundaries, covering AI agents first but aiming to stay reusable for other entity types. The first layer handles automatic advertisement and collection within each local site without requiring a particular local protocol; the second links site gateways into a federation that shares compact metadata records across administrative domains, fetching full capability documents only when needed via authenticated requests. An export-policy engine controls what metadata leaves each site, and several synchronization methods are supported. The document is informational and shows how proposals such as ACAP, Agent Directory, and ARDP can combine at boundaries.

Defines the Correctover Conformance Shape (CCS), a runtime verification framework for AI agent tool calls, with seven checks (structure, schema, latency, cost, identity, integrity, security) that produce a signed receipt carrying one of three decisions (allow/deny/escalate) and one of four execution states. This revision makes the document standalone by restating required elements of two related specs directly, corrects how implementation experience is described, and clarifies the relationship to SCITT-based agent evidence. Key management now has two trust levels, VERIFIED (offline integrity/key-consistency) and ACCEPTED (also requiring an out-of-band pinned issuer key), and the public key, its fingerprint, and the signing algorithm are now covered by the signature to block key swaps and downgrades.

Addresses AI agents calling tools on the same machine over local channels such as MCP's stdio transport, where web authorization doesn't reach and reusable credentials often sit in agent memory. Proposes a per-call proof: a Call Authority running outside the agent checks policy and issues a short-lived, single-use proof tied to the exact tool name and arguments, and the tool server asks that authority to verify it, rejecting any call whose proof is missing or invalid. Neither agent nor tool server needs to interpret the proof format, which separate profiles define, and a binding for the MCP stdio transport is included.

Defines a hybrid key-exchange method for TLS 1.3 pairing the post-quantum ML-KEM-1024 algorithm with the classical X448 curve, as a stronger option than X25519MLKEM768 reaching the security level of AES-256 and ChaCha20. Its larger security margin is meant to guard against future cryptanalysis, misuse, and implementation mistakes, and X448 is faster and more robust than similar-level P-curves. A FIPS-validated implementation keeps the ML-KEM-1024 component FIPS-validated. The method targets TLS 1.3, DTLS 1.3, and QUIC.

Defines an HTTP caching extension that changes how a URL's query string affects whether a stored response can be reused. A new response header, No-Vary-Search, lets a server declare which query parameters don't change the returned content, so caches can ignore those parts when matching a stored copy. As a result, URLs that differ only in ignorable parameters can share one cached response, increasing cache hits.

Describes an add-on for Media over QUIC Transport (MOQT) that lets a subscriber cap how many data streams and how many bytes a publisher may send for a single subscription. Endpoints enable it during connection setup and can set initial limits there. The extension defines parameters for communicating limits, messages for granting more allowance and for reporting when a sender is held back, and a session error code for when a limit is violated.

Defines a YANG data model for managing network incidents end to end, where an incident is a problem that disrupts or degrades a network service. The model gives systems a common format for reporting incidents and supports diagnosing root causes. By grouping related issues into a single incident it aims to cut the number of troubleshooting tickets operators handle, resolve problems faster, and protect service health.

Observes that network systems have traditionally needed a pre-agreed data model, or a richer shared ontology, both requiring up-front negotiation. Proposes that once systems can reason, each side converts its own data into a complete semantic model and two such models are reconciled on the spot without any pre-agreed model, via a 'lift' that builds a model from one system's data and a reconciliation that bridges two models. It emphasizes pragmatics, the context-of-use layer schemas rarely capture, examines thin shared references, and reports an empirical study across four network-management scenarios using several AI-agent families. It is an IRTF NMRG research contribution and does not represent IETF or IRTF consensus.

Defines a Traceroute Extension Block (TREB) for Bundle Protocol Version 7. As a bundle travels, each participating node adds a record of its own identity, the node it received from, the next node chosen, the event time, and link details, and the block is copied into status reports that carry the trace back to the sender. Use is optional and aimed at special diagnostic bundles on scheduled, bandwidth-limited paths: each forwarding report returns one node's record while the delivery report returns the full path plus a return path, enabling a round-trip trace. Status report generation remains governed by RFC 9171.

By its name alone the subject of this individual submission ('aer1', revision 12) is not clear, and the official abstract could not be retrieved on this run (the fast page returned 404). This description is therefore approximate; the document's scope will be grounded from its abstract on the next run.

Proposes a delivery time limit for media Objects over Media over QUIC Transport (MOQT). Today the delivery timer restarts at each relay, so an Object can get a fresh budget at every hop; instead the proposal uses each Object's timestamp so the timeout counts delay accumulated across a chain of relays. The first Object sets a local reference point, and later Objects that fall too far behind that timeline are no longer forwarded. The scheme works without synchronized clocks.

Introduces OpenA2A AIP, an open standard for creating, managing, and verifying cryptographic identities for AI agents spreading across browsers, cloud platforms, and enterprises, aiming to answer which agent is present, what it may do, and whether it should be trusted. Central features are an independently checkable trust score and a portable signed credential using both classical and post-quantum signatures, plus a tamper-evident public log of identity and credential issuance, decentralized identifiers scoped to each provider or the ecosystem, and a vocabulary for agent capabilities. It also covers challenge-response verification, governance, lifecycle, and audit logging, and notes it is meant to complement existing standards.

Addresses a gap in RFC 9681's IS-IS fast-flooding mechanism, which caps unacknowledged LSPs via a receive window but leaves sender-side congestion control optional, so a sender can keep retransmitting at full speed and sustain a retransmission storm. Proposes that a sender lower its outstanding-LSP limit by a meaningful fraction whenever retransmissions to a neighbour exceed a threshold within a recent window, and adds a single-bit Retransmission (RTX) flag a node sets only to report its own retransmissions so a neighbour can shrink its advertised window and the adjacency settles at a sustainable rate. Multiple versions of the same LSP fragment count as one pending LSP for the relevant thresholds.

Proposes HMTP, a standard for sending and receiving email over HTTPS, covering both message submission from a user's client to their provider and server-to-server relay. Messages keep their standard email format, wrapped in a JSON package holding delivery details, and large files can be sent as links with fingerprints so recipients can confirm integrity. Each request is signed with domain-published keys (like DKIM), so receivers verify the sender without relying on IP addresses, and servers find each other's HMTP endpoints via DNS. For adoption, a server can fall back to SMTP when the peer doesn't support HMTP.

Reports on the current state of mutual-TLS (mTLS) client authentication in the Extensible Provisioning Protocol (EPP). A policy change on 15 June 2025 led some Certificate Authorities to change the client certificates they publish, affecting how EPP clients prove identity to a server. The document describes the problem and outlines options for handling its operational effects, aiming to keep mTLS authentication working for EPP; specific fixes are covered in the body rather than the abstract.

Defines VXLAN, a way to build virtual overlay networks inside multi-tenant data centers for both cloud providers and enterprise-run data centers, and replaces RFC 7348 which had described already-deployed VXLAN. The specification now moves to the IETF standards track, enabling registered extensions to the VXLAN header via IANA. The packet format and processing rules are unchanged, keeping it fully compatible with RFC 7348.

Defines an experimental profile for controlling how confidently a machine-generated claim may be stated, with the core rule that a claim must not be presented as more certain than its validated, traceable evidence supports. It separates whether a statement matches reality from whether the system honestly conveys how well it is supported, making no promises that sources are accurate or reasoning never fails. It provides labels marking a statement as proved, observed, sourced, interpretive, a rule, a prediction, or unresolved, and is meant to work alongside the separate QIK-VRT EFFECT_ACK mechanism (which authorizes downstream actions) without changing its wire format.

Notes that data sent over HTTP can be interrupted when a request is cancelled or a connection drops, and that if the receiver can report how much it already processed the sender can resume rather than resend everything. HTTP already supports this for server-to-client downloads via range requests; this document defines a comparable mechanism for client-to-server uploads, letting uploads recover from interruptions using HTTP.

Specifies how to combine ML-KEM, NIST's post-quantum key-establishment method, with established methods RSA-OAEP, ECDH, X25519, and X448. The composite combinations are designed to fit common security best practices and government requirements and apply to any system using X.509/PKIX certificate formats that supports ML-KEM. The main goal is extra protection in case ML-KEM is broken or contains a serious implementation bug.

Reserves entries in several OpenPGP registries so they can be used in interoperability testing, following the same idea as GREASE in TLS. The goal is to help OpenPGP software stay compatible with future protocol changes by regularly exercising unknown-value handling.

Defines a YANG extension that tags YANG modules, submodules, and packages with a version number based on an extended form of semantic versioning, and gives guidelines for applying those rules to revisions. A second extension lets authors constrain module imports by these version numbers. The document updates RFC 7950 (YANG 1.1), RFC 9907 (YANG author guidelines), and RFC 8525 (YANG Library).

Builds on RFC 5837, which lets ICMP error messages carry details identifying the interfaces along a path (useful for traceroute when an interface lacks a unique IP address). This document proposes a similar extension for identifying nodes (the devices themselves), allowing a node to be identified by a unique IP address, a text name, or both. It matters when nodes can't be distinguished by IP address alone, such as a network where every interface uses IPv6 even for IPv4 routes.

By its name this individual submission concerns a 'helm protocol' and 'tttps', which probably denotes a transport- or TLS-related protocol variant, but the specifics cannot be determined from the name alone. The official abstract could not be retrieved on this run (the fast page returned 404), so this description is approximate and will be grounded from the abstract on the next run.

Defines what terms like 'IPv6-Only' and 'IPv6-Mostly' mean, to prevent confusion when they appear in IETF and other documents. The authors want a reference to 'IPv6-Only' to describe what a network actually uses within a given scope, reflecting the functionality in operation rather than merely which protocols are installed. The specific definitions appear later in the document.

Argues that protocols and extensions work best when operations and management are considered early, since retrofitting them is harder, and gives authors and reviewers guidance on which operational and management topics to address. It replaces RFC 5706 with updated techniques and updates RFC 2360 so a mandatory MIB is no longer required. New RFCs defining protocols or extensions, including YANG models, must include an 'Operational Considerations' section, or state that there are none.

Notes that agentic AI systems act for people and pass tasks through chains of agents, yet there is no standard way to record who authorized an agent, what was allowed, and how authority was delegated, or to check that record without a central registry. Defines HDP version 0.1, a lightweight token format linking a human's authorization to a session, with each delegation step added as a signed entry in an append-only chain that an auditor can verify using only the issuer's public key, without network calls or outside trust anchors. The content can also ride in UCAN or ZCAP-LD metadata. HDP is explicitly not an authorization protocol: a token grants no authority and must not drive access decisions.

Observes that ACME verifies domain control (via DNS or web resources) and identifies requesters by account keys, which fits poorly with automated workloads in CI pipelines, container orchestrators, and cloud platforms that hold no long-term secrets and can't publish DNS records. Those workloads can obtain short-lived signed identity tokens from their platforms, so the document defines a new ACME challenge type, wif-01, that approves a certificate request based on such a token bound to the account key and authorization through its audience field. Because identity is checked per authorization, workloads avoid External Account Binding and need no pre-shared secret, working within ACME's existing extension points.

Describes the 11-Stage AI Visibility Lifecycle, an analytical model of how AI systems find, understand, trust, and surface websites to people, grouped into AI Comprehension (stages 1-5), Trust Establishment (stages 6-8), and Human Visibility (stages 9-11). It is not strictly sequential: stages 1 and 2 always come first for a page, but later stages are weighed together and a domain can advance on several at once. The numbers are illustrative estimates and the described AI behaviours are inferred from observation rather than documented internals. Its main contribution is a shared vocabulary and dependency map separating being crawlable from being visible; the author notes figures and some mechanisms are provisional and need testing.

Defines an optional new IS-IS packet type for database synchronization, the Aggregated SNP Hash (ASH), which bundles the usual SNP summaries into a Merkle-tree-like structure where summaries cover groups of summaries. This helps routers synchronize large link-state databases and adjacency information more quickly and reduces the routine CSNP traffic exchanged during normal operation. ASH comes in two forms, Complete ASH (CASH) and Partial ASH (PASH), paralleling the existing CSNP and PSNP packets.

A practical guide for running mobile networks on IPv6 alone, focused on 5G and using the 464XLAT translation method so devices can still reach IPv4-only services. It explains core 5G concepts and architecture, the ways operators can configure IPv6-only mobile networks, and the practical problems operators are likely to face during deployment.

Observes that LISP routers (ETRs) and Map-Servers currently exchange information over unreliable UDP and so resend the same data on a timer to keep state current, placing a steady load on both devices that scales poorly as LISP carries more information. Proposes using a reliable transport protocol for this communication instead, removing the need for periodic repeated messages while providing delivery guarantees, flow control, and detection of an unavailable endpoint.