Describes an approach to information distribution and subscription within an autonomic network using GRASP (the GeneRic Autonomic Signaling Protocol). Rather than extending GRASP itself, the mechanism is layered on top of existing GRASP features to build a distribution-and-subscription service, letting autonomic nodes publish information and subscribe to updates using capabilities the protocol already provides.
Explains how devices already admitted to an Autonomic Control Plane (ACP) can use MQTT, the OASIS publish/subscribe messaging standard, to exchange arbitrary information among themselves. The design requires an MQTT message broker whose location can be advertised through the GRASP flooding mechanism, so nodes can discover the broker and participate in pub/sub messaging over the secured ACP.
Specifies SPA-based SAVNET, an intra-domain Source Address Validation (SAV) mechanism in which routers inside a network share Source Prefix Advertisement (SPA) information. Boundary routers combine routing information and SPA information to automatically generate accurate SAV rules, forming a validation boundary that checks the source addresses of packets entering the network. The document describes how to carry and use SPA information with the OSPF and IS-IS link-state IGPs.
Defines a BGP-based method for intra-domain Source Address Validation at the network edge. Boundary routers automatically build accurate SAV filtering rules from their routing information together with additional SAV-specific data, and those rules verify whether the source address of each incoming packet is valid. The draft specifies BGP extensions that let boundary routers distribute the SAV-specific information needed to construct the rules.
Defines the Morning Brief, a signed, short-lived situational-awareness payload exchanged between organisations, their agents, and peer agents under an Identity Accord. Each brief bundles signals, escalations, decisions, and optional commerce quotes from one identity to another, and every signal is labelled by how it was gathered (self-reported, passively aggregated, or passively observed), with the last category barred on the wire and rejected at parse time. Briefs are signed, bound to the issuer's identity, expire after 24 hours by default, and advertise a DNS-discoverable revocation endpoint. Recipients present a token bounding what they may receive, which the issuer enforces before sending. Only the wire format is in scope.
Proposes a wire-level annotation grammar for large-language-model output. Each assertion can carry a label naming the external source (drawn from a fixed, versioned list) that corroborated it, or one of two terminal states: unverified inference, or decayed to uncertainty. The format does not dictate how an inference system behaves; it simply lets a relying party see what the system actually consulted when producing a statement. The document is Informational.
An Informational memo describing a coordination anti-pattern in cross-tool agentic systems, where multiple sessions of a single identity-bound principal coordinate by exchanging a shared wire format. It proposes an alternative in which sessions instead observe byproducts of each other's normal activity, so no coordination wire format needs to be negotiated. The memo proposes no standard protocol element, explains what should not be standardised for this problem and why, and names existing substrate mechanisms that fill the role.
Describes a compute-location gate: a wire-layer negotiation in which a client and an identity-inference server agree, before any inference runs, on where it will run, based on the provenance class of the input signal. Active signals (principal-initiated) allow server-side computation with server storage; passive-aggregate signals (from a cohort above a minimum size) allow server-side computation yielding only population-level output; passive-individual signals (observed without the principal's initiation) stay on the observing device and are never sent to a server. Enforcement is by consent-class matching and wire-layer refusal, not cryptographic proof. Informational.
Describes a two-layer federated reference architecture for initial discovery of minimum public discovery information about AI resources, developed in the DAWN working group. The Local Discovery Plane handles zero-configuration advertisement within a single site without a specific link-local protocol, and the Federation Plane lets site gateways exchange integrity-protected Federation Metadata Records across administrative domains, emphasising data sovereignty through an Export Policy Engine. DNS delegation (Mode A) with DNSSEC is preferred, with non-DNS options available. Informational; it excludes open-ended search, semantic matchmaking, ranking, and capability catalogs.
Proposes extensions to BGP FlowSpec for disseminating Source Address Validation (SAV) rules. Noting that BGP FlowSpec already reuses BGP routes to distribute infrastructure and propagate traffic-flow information with filtering actions, the draft adds the ability to carry SAV rules so they can be distributed and applied across the network using the existing FlowSpec machinery.
Addresses AI agents that call local tool servers on the same host (for example via MCP's stdio transport), where channels sit outside HTTP-based authorization and a tool server cannot tell whether a call passed a policy check. It proposes a per-call proof: a separate Call Authority evaluates policy and issues a short-lived, single-use proof bound to the exact tool name and arguments via a canonical digest. The tool server rejects any call whose proof is missing or fails verification. The proof format is opaque to agent and server and defined in separate evidence-type profiles, with a binding for MCP stdio.
Provides a systematic analysis of emerging agent-based authorization use cases. It groups the use cases into scenarios, lists each scenario's authorization requirements, and compares them against OAuth 2.0 (RFC 6749) and its common extensions. The goal is to identify fundamental gaps and lay groundwork for new OAuth Working Group extensions that support a more secure and interoperable ecosystem for agent-based systems.
Profiles a draft on signed decision receipts for machine-to-machine access control, adapting it for use as card-scheme dispute evidence. It explains how a receipt is populated for payment disputes and adds four evidence layers: card-scheme dispute categories, agent-specific anchors, a record of the money that actually moved, and a custody wrapper aligned with U.S. evidence rules. It documents one vendor's implementation rather than defining a new protocol, and claims no IETF consensus or standardization.
The official abstract could not be retrieved from the fast www.ietf.org source on this run, so this description is approximate and will be grounded on the next run. By its filename the document is an individual submission (revision -16) under the identifier "aer1"; its precise scope could not be confirmed from the index alone. Refer to the linked draft for the authoritative abstract.
Notes that many protocols add header fields to packets on entry to a network domain and remove them on exit, for example Tenant IDs in multi-tenant networks, ingress port identifiers, and other identity or handling directives, so a packet can carry information it would otherwise lack inside the domain. The main concern is privacy: such fields can reveal more precisely where traffic originates and how it is to be handled. The document provides security considerations for including these fields in packets.
The official abstract could not be retrieved from the fast www.ietf.org source on this run, so this description is approximate and will be grounded on the next run. By its name, this individual draft (new, -00) concerns provenance of rules in agent systems, i.e. recording where an agent's governing rules came from. Refer to the linked draft for the authoritative abstract.
Specifies two DNS components for storing information about connection tunneling or encapsulation: a Service Binding (SVCB) service parameter type and a new DNS resource record type. Together they let tunneling details be published in, and retrieved from, the DNS.
The official abstract could not be retrieved from the fast www.ietf.org source on this run, so this description is approximate and will be grounded on the next run. By its name, this individual draft (revision -07) concerns an audit trail for agent systems, i.e. recording the actions of AI agents for later inspection. Refer to the linked draft for the authoritative abstract.
Explains how a Signed Evaluation Receipt, recording one evaluation result as signed payload bytes B, can be cited or carried by existing envelopes: an envelope can reference it by the SHA-256 digest of B, and a DSSE envelope can carry B with the receipt's own signature. It maps receipt parts onto DSSE, the in-toto test-result Statement, and the SLSA Verification Summary Attestation, covers a SCITT Signed Statement that names a receipt by digest, declares the receipt digest as an artifact type, and relates a receipt to a pre-run criteria set in PRML. It defines no new format and no new predicate.
Proposes requirements and test cases for identifying SCITT Signed Statements, separating reference matching from retrieving a statement, verifying its signature, authorizing its issuer, and proving registration. It also examines how different identification schemes handle changes to statement encodings and registration context. The draft is offered as input to technical discussion rather than working-group consensus, and does not define a common reference format, assign a COSE header parameter, or establish a relationship vocabulary.
Defines a Signed Evaluation Receipt: a JSON object recording one evaluation result, namely whether a stated threshold was met for a stated metric of a stated test suite, signed with Ed25519 over a domain-separated encoding of its canonical bytes. The draft stresses the receipt's limits: it identifies who signed the recorded bytes but not whether the recorded result is true, and it does not prove registration with a Transparency Service. Under SCITT, transparency additionally requires registering a Signed Statement carrying the receipt or its digest and verifying the resulting SCITT Receipt, which is out of scope here.
Observes that agent systems can send control instructions (stop, suspend, revoke, override) across system and administrative boundaries, and that a record of a control being decided or dispatched does not prove every enforcement point received or applied it, nor does a missing acknowledgement prove non-delivery. It sets out format-independent evidence requirements that keep these facts separate (issuer-side emission, target resolution, receiver-side observation, enforcement outcome, observed effect), making each instruction-target pair the unit of reconciliation. It defines no receipt format, wire protocol, authorization system, policy language, transparency service, or audit regime.
Proposes an IPv4 option that carries a flow label compatible with the IPv6 flow label. The authors argue the option is needed to simplify the use of Integrated Services (IntServ) and to improve interoperability with IPv6, giving IPv4 packets a flow-identification field analogous to the one available in IPv6.
Describes a design for inter-domain multicast overlays built on the Locator/ID Separation Protocol (LISP), explaining how such overlays operate over both multicast and unicast underlays. It describes a signal-based approach using PIM that programs LISP encapsulators with a replication list within a locator-set, where the list can mix multicast and unicast locators. Once approved, the document will obsolete RFC 6831.
Presents system and Radio Access Network (RAN) requirements for 3GPP AI/ML and AI Agents in 6G networks. It states the requirements depend on the type of agent and its application area, and describes requirements for several agent types as well as RAN requirements for 6G, including those related to AI. It also covers implementation aspects: how APIs can be discovered, how agents can be found across and within domains using DNS lookups, and how agent protocols can be designed.
Defines a new QUIC transport frame that lets consenting endpoints share information about their congestion-control state for a range of purposes. It also allows a peer to return an endpoint's own congestion-control state back to it, so the endpoint can consider that state at the start of a later connection.
Introduces the Sovereign AI Horizontal Memory (SAIHM) protocol, a memory layer for AI agents in which an operator stores an agent's memory as encrypted cells tied to a post-quantum holder identity, logs each state change in a public audit log, and supports revocable sharing contracts. It defines an erasure operation with two erasure classes to help operators meet GDPR Article 17. This revision covers the custodial profile, where the operator holds the keys and performs all cryptography while the agent-side client holds no keys, uses only TLS, keeps no local store, connects over HTTPS, and exposes the protocol as eight MCP tools.
Extends the resumable upload protocol so a client can send different parts of one upload at the same time, instead of sequentially. Servers and intermediaries can cap both the number of concurrent requests and the amount of out-of-order data they accept. It explains how intermediaries that inspect uploads (for example for data-loss prevention or malware scanning) can keep doing so, and how the extension works over HTTP/1.1, HTTP/2, and HTTP/3.
Covers changes from Unicode 12.0.0 to Unicode 18.0.0 that affect the values IDNA2008 produces. It identifies newly added code points needing individual examination and asks IANA to mark them "UNDER REVIEW" in its tables until that examination completes, as RFC 8753 permits, and supplies the table IANA needs to align its database with Unicode 18.0.0. All values were computed directly from the released Unicode Character Database files by applying the RFC 5892 Section 3 algorithm, with Appendix G comparing the results against the Unicode Consortium's published derivation.
Notes that OAuth metadata lists supported scopes and RAR types but does not describe the authority those identifiers represent. It proposes authority-metadata extensions for Protected Resource Metadata and Authorization Server Metadata: protected resources could describe the authority behind their scopes and RAR types, define reusable access-requirement profiles, and point to lower-authority alternatives, while authorization servers could publish a simpler description of their authority vocabulary. This metadata could inform authentication and authorization decisions before tokens are issued or exchanged, and could support authority attenuation.
Defines the P10 profile, a way for third parties to verify a "NotDemonstrated" result with the reason "underdetermined." A conforming statement includes two canonical worlds that both fit the same closed evidence set yet assign different values to the same frozen claim. The witness result is checked against the profile's committed semantics and bound into a SCITT Transparent Statement using an in-toto Statement v1 predicate. The result shows underdetermination only relative to the declared profile; it does not identify which world is actual or show either claim value is true.
Introduces the Sovereign AI Horizontal Memory (SAIHM) protocol, a memory layer for AI agents in which an operator stores an agent's memory as encrypted cells tied to a post-quantum holder identity, logs each state change in a public audit log, and supports revocable sharing contracts. It defines an erasure operation with two erasure classes to help operators meet GDPR Article 17. This revision covers the custodial profile, where the operator holds the keys and performs all cryptography while the agent-side client holds no keys, uses only TLS, keeps no local store, connects over HTTPS, and exposes the protocol as eight MCP tools.
Defines an extension to the Dynamic Link Exchange Protocol (DLEP) for reporting how much of a radio channel is in use, letting a radio make channel-utilization information available to its DLEP peer.
Defines an extension to the Dynamic Link Exchange Protocol (DLEP) that reports which frequency bands a radio uses, making that band information available to the radio's DLEP peer.
Extends the Dynamic Link Exchange Protocol (DLEP) to make the quality of incoming radio signals available to DLEP peers, so routers can take received-signal quality into account.
Describes how BGP Flow Specification (FlowSpec) rules can steer matching traffic into Segment Routing (SR) Policies. It defines two modes: Mode 1 provides transport steering only, in SR-MPLS and SRv6 networks, signaled with BGP Extended Communities; Mode 2, for SRv6 networks, adds an egress service action performed at the egress node, signaled with the BGP Prefix-SID Attribute.
Argues that large language models trained on huge datasets, with billions or trillions of parameters, require thousands or tens of thousands of GPUs in a single cluster, so three- or five-stage CLOS networks are common in AI data centers and a non-blocking fabric grows more important as models scale. It concludes that adaptive routing is needed to spread traffic toward one destination across several equal-cost paths, using information about each path's capacity.
Observes that delegation-chain specifications describe the shape of conveyed authority but leave the verifier's side underspecified, so two verifiers can check the same chain, both succeed, and still enforce different policies. It specifies what a verifier must do: the explicit inputs evaluation depends on, how those inputs should behave when their sources are stale or unavailable, and four rules that keep evaluation fail-closed. The rules are drawn from the GAL and PTC specifications and a public reference implementation.
Sets architectural requirements for agent systems that rely on evidence, provenance, model outputs, or other machine-readable claims before taking consequential actions. It requires implementations to keep five concerns separate: artifact integrity, provenance, evidence qualification, authorization, and execution. It further requires unresolved required evidence to be shown explicitly rather than hidden, and issued evaluation records to be preserved in append-only form. The document defines no new transport protocol and requests no IANA registrations.
Describes a transport-neutral model covering how distinct representations can correspond within defined bounds, and setting explicit promotion boundaries between observation, evidence, authorization, and execution, so that each stage is not conflated with the next.
Describes AID-1, a provider-independent architecture for verifying the identity, delegation, authorization, action evidence, execution attestation, provenance, and governance information tied to AI systems and the actions they mediate. It keeps signature validity, identity binding, authority, execution evidence, provenance, and downstream admissibility as separate determinations, so success in one does not imply success in another. The specification covers trust domains, signed envelopes, external key resolution, delegation credentials, authorization decisions, temporal and revocation checks, replay protection, and typed provenance claims, and defines a deterministic verification order yielding ALLOW, DENY, or INDETERMINATE.
Defines the Agent Delegation Receipt, a signed record of what a human principal permitted an AI agent to do, for how long, and under which instructions, including a scope, hard boundaries, a validity window, and a hash of the operator's instructions, all signed by the authorizing key. Earlier revisions replaced a custom signature format with JWS (over JCS-canonicalized JSON) and COSE_Sign1 profiles, scope narrowing across multi-agent chains, and a cascading revocation registry. Revision -12 settles unknown-field handling: authorization fields are closed so verifiers reject unrecognized top-level fields, a "metadata" object is the extension point, and one-time-use receipts are denied without a durable consumption store.
Recalls that BGP FlowSpec (RFC 8955 and RFC 8956) distributes traffic filter policies (filters and actions) over BGP, already used for DoS mitigation, traffic filtering in BGP/MPLS VPNs, centralized firewall control, and SFC traffic insertion. It argues FlowSpec can also be readily applied to network congestion management and describes how to use it for that purpose.
Specifies the Vaara Receipt format (vaara.receipt/v1), a signed record tying a decision about an autonomous action to the evidence behind it, with optional external timestamp anchors. Using JSON Canonicalization Scheme encoding, a third party can recompute digests and verify public-key signatures without issuer code; a decision receipt and its execution receipt form a recomputable pair via a back link. Trust does not depend on a single root, the receipt can be verified with or without a hardware TEE, and it can be re-expressed as an IETF RATS Entity Attestation Result. The format is already deployed, with public conformance vectors and standalone checkers.
Offers a network-operator framework for building and running IPv6-only core networks that span several interconnected Autonomous Systems. To carry remaining IPv4 traffic it proposes stateless IPv4/IPv6 address mapping as the basis for IPv4-as-a-Service, translating packets at the network edge so they cross the IPv6-only core without per-flow state or conversion gateways in the data path. Intended as a problem statement, guidance, and requirements rather than a protocol specification, it addresses applicability and trust boundaries, IPv6 mapping-prefix allocation options, and operational, manageability, and security considerations.
Argues that agent protocols often need a decision-time input before acting (issuer standing, key availability, revocation status, a downstream outcome), and that receiving a negative answer is different from being unable to establish the input at all; conflating the two distorts retries, alert routing, audit interpretation, incident ownership, and reconstruction. It sets encoding-independent requirements for preserving evaluation state across agent-protocol boundaries, covering the input's value, whether it was established at decision time, source freshness, and whether the policy decision was actually evaluated. A later revision will define concrete representations.
The official abstract could not be retrieved from the fast www.ietf.org source on this run, so this description is approximate and will be grounded on the next run. By its name, this new individual draft (-00) concerns multicast in Delay-Tolerant Networking (DTN), i.e. delivering bundles to multiple destinations over disruption-tolerant links. Refer to the linked draft for the authoritative abstract.
Describes an experimental email header field, Conversation-ID, that participating senders can use to group messages into a conversation alongside Message-ID, In-Reply-To, and References. The identifier is separate from any single message's Message-ID, and its value depends on senders including it and intermediaries preserving it in transit. The draft calls it "an untrusted grouping hint" with no authentication or authorization, specifies the field's syntax and processing rules, and outlines an experiment comparing it with existing correlation methods.
Explains that incomplete deployment of Route Origin Validation (ROV) can make some BGP hijacks less visible in the control plane while they still divert traffic. The document describes the mechanism behind this, defines the threat, analyzes a real-world incident that illustrates it, and discusses possible countermeasures.
Explains how a Signed Evaluation Receipt, recording one evaluation result as signed payload bytes B, can be cited or carried by existing envelopes: an envelope can reference it by the SHA-256 digest of B, and a DSSE envelope can carry B with the receipt's own signature. It maps receipt parts onto DSSE, the in-toto test-result Statement, and the SLSA Verification Summary Attestation, covers a SCITT Signed Statement that names a receipt by digest, declares the receipt digest as an artifact type, and relates a receipt to a pre-run criteria set in PRML. It defines no new format and no new predicate.
Proposes requirements and test cases for identifying SCITT Signed Statements, separating reference matching from retrieving a statement, verifying its signature, authorizing its issuer, and proving registration. It also examines how different identification schemes handle changes to statement encodings and registration context. The draft is offered as input to technical discussion rather than working-group consensus, and does not define a common reference format, assign a COSE header parameter, or establish a relationship vocabulary.
Defines a Signed Evaluation Receipt: a JSON object recording one evaluation result, namely whether a stated threshold was met for a stated metric of a stated test suite, signed with Ed25519 over a domain-separated encoding of its canonical bytes. The draft stresses the receipt's limits: it identifies who signed the recorded bytes but not whether the recorded result is true, and it does not prove registration with a Transparency Service. Under SCITT, transparency additionally requires registering a Signed Statement carrying the receipt or its digest and verifying the resulting SCITT Receipt, which is out of scope here.
Specifies the Agent Identity Protocol (AIP), which carries delegated authority along a chain of AI agents so that a relying party, using the token and the issuer's public key, can confirm who granted the authority, which agents it passed through, and that no hop granted itself more than it received. AIP targets hops without an authorization server (such as one agent delegating to a subagent) and complements the AIMS framework for hops that do have one. It defines an encoding-independent delegation-chain model with a single attenuation rule, a normative verification procedure, two encodings (chained JWTs and a Biscuit token with Datalog policy), and bindings for MCP, A2A, and generic HTTP.
Describes how an AI agent runtime, tied to a principal's identity handle, fetches its governing policy at session start: it finds the target organisational identity substrate via a manifest source, then retrieves four policy components (a handbook, a pointer to an SOP registry, an enforcement-gate specification, and an audit-signal endpoint), enforcing them as constraints on tool calls and sending audit signals back to the substrate. Policy is supplied in the same step as identification. If a principal belongs to several substrates, their policy stacks combine deterministically, and unresolvable conflicts suspend the action, settled only through the Identity Accord process. Informational.
Proposes a way to reference identities alongside the existing "~handle" taxonomy: a pronoun is a session-scoped reference the client resolves to a concrete handle before any cryptographic, DNS, or federation operation. The existing entity classes (Sovereign, Bot, Instrument) are unchanged, and a separate Absolute-versus-Pronoun distinction is added. Pronouns must never appear in capability tokens, DNS records, Accord signatures, or inter-organisational payloads. One pronoun, "~org" (the organisation bound to the caller's current session), is defined now, with a relative-path pronoun grammar left as non-normative future work.
Defines the agent-channel frame, an application-layer message format for exchanging short structured messages among an identity-bound principal's concurrent agent sessions and members of an organisational identity substrate. Each frame carries a closed-catalogue kind, a kind-specific structured payload, an identity attribution pair, and an inline provenance block. A sender names a recipient scope rather than a single endpoint, expanded at delivery time against the recipient's subscriptions; recipients receive frames over a per-handle Server-Sent Events stream and can narrow them with a filter. Frames are ephemeral; persistence for replay is out of scope.
Proposes an IANA registry for Model Context Protocol (MCP) tool surface names, the identifiers clients use to invoke typed capabilities on MCP servers, setting up the registry so other specifications can register names without restating its structure or procedure. Registration uses RFC 8126's Specification Required policy with a Designated Expert pool. The initial contents are four surface names from a companion policy-provision specification (one Active, three Provisional), and the draft recommends, but does not require, vendor prefixes for names.
Describes a profile for registering an autonomous agent with no human or organisational principal at the root of its delegation chain. Registration creates a payee record for the agent but does not make the agent a principal: value credited to that record goes to a sponsoring person who accepted the agent, or, absent a sponsor, is held in trust until an operator-defined condition is met. Admission of the agent's signed statement depends on settling an HTTP 402 (Payment Required) challenge, and when payment is authoritative to admission the payment proof is committed to the service's verifiable data structure so admission stays deterministic and replayable. Informational.
Proposes an extension to HTTP-native agent payment protocols for reading identity attributes about people. When an agent pays for such a read, the request must reference a scoped, revocable consent grant issued by the person, and the payment's settlement instruction must name that person as a beneficiary whose share exceeds the combined shares of all other parties. The extension sits atop an identity-attestation layer and a payment layer, adding consent capture at disclosure and settlement to the data subject through three additions (an advertisement, a consent-grant reference, and a subject-beneficiary settlement instruction). It is settlement-network and attestation-format independent. Informational.
The document has no Abstract section on the fast www.ietf.org source, so this description is written from its name and is approximate; it will be grounded on the next run. By its title, the draft concerns optimizations to PIM multicast snooping, i.e. improving how switches snoop PIM signaling to constrain multicast forwarding. Refer to the linked draft for the authoritative text.
Introduces EDHOC-AKA, an Authentication and Key Agreement method for the EDHOC key-exchange protocol, defined as a profile of the EDHOC Pre-Shared Key (PSK) method. It lets a resource-constrained device with a mobile subscription (SIM/UICC) set up a forward-secret secure session with an application server: a 3GPP AKA challenge-response runs inside standard EDHOC messages via External Authorization Data fields, with the EDHOC Responder relaying AKA messages to the mobile core network. After a successful exchange a session-specific key becomes the PSK for EDHOC-PSK. It targets Non-Terrestrial Networks and massive IoT and is more compact than general frameworks like EAP.
Describes an operational approach to Computing-Aware Traffic Steering (CATS) that does not rely on the normalized scores of the CATS Metrics framework. Instead, service sites report service-oriented metrics with explicit meanings and units (such as Global Available Slots and Computing Time), and a Path Selector (C-PS) uses them directly to choose both a service instance and a network path. It explains how the metrics derive from resource and service-reference information and local policy, how the C-PS combines the Computing Service Table with the Network Service Table, and update-control and fallback mechanisms for large deployments. It complements the normalized-score method.
Describes a way to quickly protect the egress node and egress link of a Segment Routing over IPv6 (SRv6) path, since TI-LFA protects transit nodes and links but not the egress. The solution uses a Mirror SID (End.M) to redirect traffic to a protector egress when the primary egress fails. The IS-IS and OSPFv3 extensions for advertising the Mirror SID and protected locators are specified in a separate companion draft that builds on RFC 9352 and RFC 9513.
Defines GCMF, a lossless compression scheme that describes data sequences with mathematical functions and their parameters instead of storing every value. The draft covers the data format, the function types, the encoding and decoding rules, and the interoperability requirements that different implementations must meet.
Describes the Radio Image Framing Protocol (RIFP), a compact one-way protocol for sending images over low-data-rate radio links. It covers independently synchronized frames, an extensible binary header, and reassembly of fragmented data, with per-frame CRC-32 integrity and whole-image SHA-256 integrity, and optional JSON metadata plus registries for future extensions. The protocol is independent of frequency and modulation, and it defines an initial profile, rifp-cpfsk-4800, using continuous-phase binary FSK; 433.92 MHz is cited only as an example, subject to local regulations.
Defines an optional authentication extension for the Radio Image Framing Protocol (RIFP). It uses a critical header field to authenticate the compact object descriptor with HMAC-SHA-256 and a pre-shared key, where the descriptor ties the encoded image digest and decoding parameters to a session and chunk count. The extension works with RIFP's existing framing, fragmentation, retransmission, and offline IQ processing, but provides no confidentiality, no replay protection, and no public-key digital signature.
Addressed in the SPRING working group, this draft concerns using a Segment Identifier (SID) as a packet source address in Segment Routing deployments. (The official abstract could not be read in full on this run; the linked draft carries the authoritative text.)
The official abstract could not be retrieved from the fast www.ietf.org source on this run, so this description is approximate and will be grounded on the next run. By its name, this individual draft (revision -03) describes use cases and derived protocol requirements for agentic AI communication, intended to guide IETF standards work on agent-to-agent and agent-to-tool communication. Refer to the linked draft for the authoritative abstract.
Defines a "sustainability-data" well-known URI that a web origin uses to publish one JSON document with energy-consumption, carbon-footprint, and related environmental metrics for a subject (usually the origin itself). The document has a fixed path and formal schemas, links to a methodology, and can include an embedded signature, references to upstream providers' declarations, and links to third-party attestations. The metrics are self-asserted claims of the publisher. The draft registers the well-known URI and the media type application/sustainability-data+json.
Defines the Agent Delegation Receipt, a signed record of what a human principal permitted an AI agent to do, for how long, and under which instructions, including a scope, hard boundaries, a validity window, and a hash of the operator's instructions, all signed by the authorizing key. Earlier revisions replaced a custom signature format with JWS (over JCS-canonicalized JSON) and COSE_Sign1 profiles, scope narrowing across multi-agent chains, and a cascading revocation registry. Revision -12 settles unknown-field handling: authorization fields are closed so verifiers reject unrecognized top-level fields, a "metadata" object is the extension point, and one-time-use receipts are denied without a durable consumption store.
Describes a two-layer federated reference architecture for initial discovery of minimum public discovery information about AI resources, developed in the DAWN working group. The Local Discovery Plane handles zero-configuration advertisement within a single site without a specific link-local protocol, and the Federation Plane lets site gateways exchange integrity-protected Federation Metadata Records across administrative domains, emphasising data sovereignty through an Export Policy Engine. DNS delegation (Mode A) with DNSSEC is preferred, with non-DNS options available. Informational; it excludes open-ended search, semantic matchmaking, ranking, and capability catalogs.
Observes that autonomous software agents now act on the public internet without a verifiable name, a published accountable party, or a way for others to learn an operator has withdrawn one. It specifies an Agent Identity Document, a small JSON file served at a well-known URI on a hostname assigned to a single agent, and describes how an identity provider should host these names for operators without their own domain. The format separates provider-verified facts (hostname, endpoints, lifecycle status) from operator-supplied claims, publishes dated expiring checks rather than a single trust level, and requires every identity to be held by an accountable person or organisation. It documents a practice already in production.
The official abstract could not be retrieved from the fast www.ietf.org source on this run, so this description is approximate and will be grounded on the next run. By its name, this long-running individual draft (revision -16) concerns "VLSMTRP", apparently a variable-length subnet-mask transport/routing protocol. Refer to the linked draft for the authoritative abstract.