Internet-Draft MOQ Subscriber Advantage November 2026
Bisht Expires 19 May 2027 [Page]
Workgroup:
Network Working Group
Internet-Draft:
draft-altanai-moq-ai-subscriber-advantage-00
Published:
Intended Status:
Informational
Expires:
Author:
A. Bisht
Cisco

Detecting AI-Driven Subscriber Advantage in Media over QUIC (MOQ)

Abstract

This document describes subscriber behavior that may produce sustained advantages in service outcomes or disproportionate resource consumption at shared Media over QUIC (MoQ) relays. AI-assisted optimization is a motivating scenario; observations alone cannot establish whether AI was used or caused a disparity. The document outlines contextual measurement, privacy safeguards, and a proposed extension for authorization-bound resource budgets with explicit accounting and recovery behavior. It builds on existing MoQ authorization and relay denial-of-service work. This informational document defines no new MoQ messages, token claims, registries, or standardized detection algorithm.

Discussion Venues

This note is to be removed before publishing as an RFC.

Discussion of this document takes place on the Media Over QUIC Working Group mailing list (moq@ietf.org), which is archived at https://mailarchive.ietf.org/arch/browse/moq/.

Source for this draft and an issue tracker can be found at https://github.com/altanai/draft-altanai-moq-ai-subscriber-advantage.

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."

This Internet-Draft will expire on 19 May 2027.

▲

Table of Contents

1. Introduction

Media over QUIC Transport (MoQT) [MoQTransport] supports delivery of tracks through relays that serve multiple subscribers. Endpoints can optimize track selection, request timing, subscription filters, and relative priority. These choices may be made by an AI model, a conventional adaptive algorithm, or fixed application rules.

Operators may wish to identify sustained differences in resource use or service outcomes among comparable subscribers. A disparity can be useful for capacity planning, service assurance, or a review of relay policy. It is not by itself evidence of misconduct, an unfair policy, or use of AI. Higher quality can be a legitimate result of a better path, authorized service treatment, or more efficient application behavior.

This document develops two related questions: how to observe such patterns while limiting subscriber data exposure, and how a future resource-budget profile could associate explicit usage limits with existing authorization mechanisms. The latter extends the original detection scope and is presented as design considerations for further work, rather than a completed token or enforcement specification.

2. Terminology and Scope

A subscriber endpoint can hold multiple subscriptions. A subscription is a relationship for receiving one track; it is distinct from a MoQT session, which can carry multiple subscriptions and fetches. Separate subscriber endpoints commonly use separate sessions. The object and session model is defined by [MoQTransport].

For this document, a comparison cohort is a group of observations matched for content, service policy, observation interval, and relevant path conditions. A candidate advantage is a sustained disparity in resource use or delivery outcomes within such a cohort. The designation identifies a measurement for review and does not establish its cause.

A resource budget is a proposed policy describing an allowed amount, rate, or concurrency of a defined resource. An accounting domain is the set of sessions and relay instances over which that budget applies. An accounting reference identifies the shared usage state within that domain and validity period; it need not expose an account identity to an analytics system.

In scope are observable signals, interpretation of those signals, privacy-conscious processing, resource-budget design considerations, and relationships to existing MoQ and SIP mechanisms. Out of scope are AI-versus-human attestation, universal definitions of fairness, monetary penalties, standardized machine-learning algorithms or thresholds, and new token encodings or protocol error codes. The illustrative policies below are not interoperability requirements.

3. Motivation and Problem Statement

Shared relay resources include CPU, subscription state, buffer memory, cache capacity, backend services, and upstream and downstream bandwidth. The incremental cost of a request depends on the implementation and whether suitable upstream subscriptions or cached objects already exist. Subscription aggregation can reduce upstream work. Substantial media delivery following a small request is normal protocol operation and is not sufficient evidence of abuse or harmful amplification.

Candidate concerns include:

These concerns overlap with the threat analysis and operational guidance in [RelayDoS]. This document emphasizes contextual comparison and privacy considerations, together with an exploratory connection between usage measurement and explicit resource authorization.

3.1. Priority Scope

Subscriber priority influences transmission ordering among subscriptions and fetch responses within the same MoQT session. A subscriber requesting priority 0 does not thereby acquire a protocol-defined rank above a subscriber in another session. Relays are advised not to copy downstream subscriber priorities directly to upstream subscriptions; upstream priorities depend on relay policy [MoQTransport].

Frequent priority updates may consume parsing and scheduling resources. Effects on other subscribers depend on shared-resource contention and implementation policy. A priority value, a priority change, or an improved delivery outcome alone does not demonstrate an unfair advantage. A subscriber priority adjustment also does not implement a relay-wide resource restriction across independent sessions.

4. Observable Signals and Context

The following signals are candidates rather than mandatory metrics. They are available only where a deployment can measure or attribute them. Content or timing metadata can itself be sensitive even when media payloads remain encrypted.

Table 1
Signal Interpretation and context
Subscription lifecycle activity Count creations, cancellations, and updates over a defined interval; distinguish useful adaptation, retries, and pathological churn.
Track and filter changes Compare changes relative to available content and group boundaries without retaining unnecessary viewing histories.
Priority updates Measure update frequency and work; priority values are relative within a session.
FETCH and SUBSCRIBE mix Account for historical delivery, repair, and duplicate requests; define what the byte count includes.
Resource consumption Measure attributable egress, buffered bytes, or control-processing work; identify shared costs that cannot be assigned to one subscriber.
Delivery outcomes Measure request-to-first-object or relay-observed delivery delay using defined endpoints and clocks; do not equate these with playback quality.
Cache and upstream effects Use available hit, miss, eviction, and upstream-work counters while accounting for sharing among subscribers.

[MoQMetrics] provides related measurement work. This document does not redefine its metrics or assume that every candidate signal is already standardized.

4.1. Contextual Comparison

Choose a cohort and a measurement interval before interpreting a disparity. Relevant context includes track variant and bitrate, subscription duration, publisher timing, authorized service class, cache state, path latency, congestion, and the relay's allocation policy. Normalize measurements by active time or useful delivered content where appropriate, and state the limits of attribution and sampling.

Persistence over several intervals can distinguish a sustained pattern from a short transition, but does not demonstrate causality. An analyzer should be able to return an inconclusive result when the cohort is too small or required context is missing. Controlled experiments with known client implementations can study AI-assisted behavior; relay telemetry alone cannot reliably label arbitrary endpoints as AI or non-AI.

5. Deployment Components

A possible deployment separates forwarding, budget accounting, and privacy-limited analysis. These are logical components and may be co-located. Existing token schemes provide the authorization foundation; resource-budget interpretation and accounting require further design.

  Grant issuer -- authorization --> Subscriber
                                      |
                               token + request
                                      |
                                      v
  Publisher ---- media objects ---> Relay
                                      |
                           resource-use measurements
                                      |
                                      v
                                Budget ledger
                               /             \
                    limited signals       limit decision
                           /                   \
                    Privacy guard          Relay policy
                          |                 response
                          v
                   Authorized review

The issuer grants permissions. The relay validates the presented authorization and maintains or consults usage state. The analytics path receives the minimum information needed for comparison and review, subject to the privacy process below. A comparison flag and a verified budget violation are separate inputs to operator decisions.

6. Privacy Process

The following deployment safeguards elaborate aggregation and minimal retention. They are design guidance and do not establish a formal anonymity guarantee. General privacy considerations are discussed in [RFC6973]. Related AI-preference binding work [AIPrefRealtime] can inform treatment of expressed processing preferences; a preference signal does not itself anonymize data or authorize unrelated use.

  1. Define purpose and permitted processing. Identify the measurement purpose, permitted fields, applicable preferences, and authorized recipients. Relay visibility of a field does not establish permission to retain or disclose it for every analytics purpose.

  2. Minimize collection. Prefer counters and necessary control metadata. Avoid media-payload inspection, direct account identifiers in analytics, and AI-versus-human labels inferred from traffic.

  3. Scope pseudonyms. Use short-lived identifiers specific to an analysis domain and interval. If a re-identification mapping is needed, keep it separately protected. A linkable token is pseudonymous, not anonymous, even if it does not contain a subscriber's name.

  4. Aggregate with context. Prefer interval and cohort statistics. Coarsen or suppress small groups or detailed combinations that could reveal an individual. Minimum cohort sizes and any stronger privacy guarantees need to be defined by the deployment or a future profile.

  5. Limit and audit review. Report the interval, cohort size, disparity, and measurement caveats. Restrict detailed inspection to authorized reviewers and record that access. A comparison flag alone does not establish a policy violation or justify automatic punishment.

  6. Expire detailed records. Define retention periods for events, pseudonym mappings, usage state, and review records according to their respective purposes. Delete or aggregate records when those purposes end. A future profile needs explicit retention and deletion behavior.

Budget accounting may require continuity across sessions or token renewals. That continuity should remain within the stated accounting domain and budget period. Analytics pseudonym rotation does not reset usage counters. Conversely, maintaining a budget does not justify a permanent identifier shared across unrelated operators. Privacy Pass work [PrivacyPass] is relevant to privacy-conscious authorization; maintaining shared quotas and preventing repeated issuance from resetting usage require an explicit design under the chosen token scheme.

7. Authorization-Bound Resource Budgets

MoQT already carries authorization tokens [MoQTransport]. [C4M] defines action and content scopes for Common Access Tokens and supports revalidation of ongoing authorization. The proposed direction is to specify how an authorized resource policy would be interpreted and accounted for, reusing existing token capabilities where suitable.

The following are conceptual policy elements, not registered claims:

Table 2
Policy element Design question
Service class or policy reference Which agreed policy does this label select, and which version applies?
Content and operation scope Which tracks, namespaces, and actions are permitted under existing authorization rules?
Resource allowance What concurrency, rate, bytes, or other measurable quantity is allowed?
Measurement window Is the allowance instantaneous, rolling, or tied to an explicit interval; what burst behavior is permitted?
Accounting scope Which sessions, relays, and issuer or operator domain share usage state?
Accounting reference How are related authorizations associated with one budget without revealing unnecessary identity?
Validity and renewal When does the grant expire, and how does renewal interact with remaining or already-used budget?

A subscriber cannot grant itself a more generous policy by modifying a trusted authorization. The issuer and verifier need integrity-protected claims and an agreed interpretation. Adding an arbitrary field is not sufficient: an implementation that does not understand a resource restriction must not silently treat it as an unrestricted grant. Any future encoding needs to follow the chosen token scheme's extension and validation rules. This document does not allocate such an encoding.

7.1. Usage State and Multiple Sessions

An authorization expresses a grant; it is not a mutable usage counter. For example, four authorized concurrent subscriptions require the relay to track admission and release of those subscriptions. Multiple tokens or sessions associated with the same budget must share the intended accounting state rather than each receiving the full allowance again.

A future profile needs to define atomic admission decisions, treatment of pending and paused subscriptions, cancellation, disconnects, expiry, and reclamation of abandoned reservations. If several relays enforce a shared budget, the design needs to state its consistency model and how it handles partitions or accounting-service failures. Partitioned allowances or leases are possible approaches requiring further analysis.

Revalidating an unchanged signature does not by itself reveal a quota change or update shared counters. Changes to authorization require a specified mechanism such as a stateful policy check, a replacement grant, expiry, or a supported revocation mechanism. Fresh token issuance and renewal need explicit rules so that they cannot unintentionally reset an exhausted allowance. These requirements must be reconciled with privacy and unlinkability objectives of the selected authorization scheme.

7.2. Policy Response and Recovery

A relay can enforce a clear admission or rate rule without inferring AI use, intent, or moral fault. Possible responses include rejecting an excess request, applying a temporary request-rate limit, or requiring renewed authorization where appropriate to the token scheme. Operators choose policy and thresholds; this document defines no universal penalty.

A future profile needs to describe protocol feedback, recovery conditions, restriction duration, and review of disputed measurements. It should distinguish exhausted resources or quotas from invalid authorization and use existing protocol errors according to their defined meanings. This document assigns no new error codes and does not redefine subscriber priority as a cross-session enforcement mechanism.

8. Worked Examples

All values in this section are illustrative. They are neither measured results nor standardized thresholds.

8.1. A Comparison Signal

Consider seven subscriptions to the same live track, active for ten minutes, under the same service policy and similar measured path conditions. Suppose subscriber A has 88 MB/min of relay egress compared with a peer median of 37 MB/min, and a median relay-observed delivery delay of 104 ms compared with 181 ms. The egress counter includes repeat and historical delivery as well as live objects; it does not count unique content alone.

An analyzer could report the sustained difference for review. It would still need to examine fetches, duplicate delivery, caching, publisher behavior, and measurement quality. Higher egress and lower delay do not prove that A caused another subscriber's delay, used AI, or exceeded an authorized budget. An explicit policy violation requires separate evidence against that policy.

8.2. Four Authorized Active Subscriptions

Suppose an issuer grants access to an event namespace with a policy allowing four active subscriptions across the sessions sharing one budget. The grant is valid for ten minutes. These are conceptual policy values; this document defines no claim names or serialization for them.

The relay admits four subscriptions and records four reservations. A fifth active subscription would exceed the allowance, so the relay rejects that request with appropriate existing feedback under its policy. The four admitted subscriptions remain governed by their existing terms. Repeated excess requests may trigger a documented temporary request-rate restriction. Admission can resume when a reservation is released and any temporary restriction has ended, subject to valid authorization.

The accounting reference is scoped to the operator and budget period. Detailed review is access-controlled, and retention follows the purposes specified above. Opening another session or obtaining another token for the same budget does not create another four reservations. How that association is securely established remains part of the future profile.

9. Relationship to Other Work

SIP deployments may face related questions around shared call-control or media resources. A SIP proxy and a media relay do not necessarily observe the same events. A SIP-specific design would require a separate measurement model and is not specified by this document.

10. Security and Privacy Considerations

Behavioral metadata can expose viewing patterns, correlate sessions, and support identity inference. Pseudonyms and aggregates do not eliminate those risks. Cohort size, granularity, retention, recipient access, and correlation across purposes all affect exposure. The privacy process above describes proposed safeguards, not a proof of anonymity.

A malicious client could distribute requests across sessions or tokens to avoid isolated limits. A malicious or faulty accounting component could overstate consumption and deny legitimate service. Shared ledgers and revalidation endpoints can themselves become overload targets. Budget enforcement therefore needs explicit trust boundaries, protected accounting state, bounded processing, recovery behavior, and auditable operator decisions. Local operator controls may still be needed during accounting failures or legitimate flash crowds.

An advantage signal is not an authorization credential and does not establish AI use or intent. A service-class claim does not reserve physical capacity unless the deployment separately provides such a service. This document does not change MoQT or QUIC security properties, and its examples do not authorize disclosure of subscriber identity or cross-operator reputation records.

11. IANA Considerations

This document has no IANA actions.

12. Informative References

[AIPrefRealtime]
Bisht, A., "AI Preferences for Real-Time Protocol Bindings", Work in Progress, Internet-Draft, draft-altanai-aipref-realtime-protocol-bindings-01, , <https://datatracker.ietf.org/doc/html/draft-altanai-aipref-realtime-protocol-bindings-01>.
[C4M]
Lemmons, C., Simon, G., Nandakumar, S., and W. Law, "Authorization scheme for MOQT using Common Access Tokens", Work in Progress, Internet-Draft, draft-ietf-moq-c4m-02, , <https://datatracker.ietf.org/doc/html/draft-ietf-moq-c4m-02>.
[MoQMetrics]
Jennings, C. F. and S. Nandakumar, "Metrics over MOQT", Work in Progress, Internet-Draft, draft-jennings-moq-metrics-03, , <https://datatracker.ietf.org/doc/html/draft-jennings-moq-metrics-03>.
[MoQTransport]
Nandakumar, S., Vasiliev, V., Swett, I., and A. Frindell, "Media over QUIC Transport", Work in Progress, Internet-Draft, draft-ietf-moq-transport-23, , <https://datatracker.ietf.org/doc/html/draft-ietf-moq-transport-23>.
[PrivacyPass]
Nandakumar, S., Jennings, C. F., and T. Meunier, "Privacy Pass Authentication for Media over QUIC (MoQ)", Work in Progress, Internet-Draft, draft-ietf-moq-privacy-pass-auth-04, , <https://datatracker.ietf.org/doc/html/draft-ietf-moq-privacy-pass-auth-04>.
[RelayDoS]
English, M., Pardue, L., Sharma, A., and I. Swett, "Denial-of-Service Considerations for Media over QUIC Relay Deployments", Work in Progress, Internet-Draft, draft-englishm-moq-relay-dos-02, , <https://datatracker.ietf.org/doc/html/draft-englishm-moq-relay-dos-02>.
[RFC4412]
Schulzrinne, H. and J. Polk, "Communications Resource Priority for the Session Initiation Protocol (SIP)", RFC 4412, DOI 10.17487/RFC4412, , <https://www.rfc-editor.org/rfc/rfc4412>.
[RFC6357]
Hilt, V., Noel, E., Shen, C., and A. Abdelal, "Design Considerations for Session Initiation Protocol (SIP) Overload Control", RFC 6357, DOI 10.17487/RFC6357, , <https://www.rfc-editor.org/rfc/rfc6357>.
[RFC6973]
Cooper, A., Tschofenig, H., Aboba, B., Peterson, J., Morris, J., Hansen, M., and R. Smith, "Privacy Considerations for Internet Protocols", RFC 6973, DOI 10.17487/RFC6973, , <https://www.rfc-editor.org/rfc/rfc6973>.
[RFC7339]
Gurbani, V., Ed., Hilt, V., and H. Schulzrinne, "Session Initiation Protocol (SIP) Overload Control", RFC 7339, DOI 10.17487/RFC7339, , <https://www.rfc-editor.org/rfc/rfc7339>.
[RFC7415]
Noel, E. and P. Williams, "Session Initiation Protocol (SIP) Rate Control", RFC 7415, DOI 10.17487/RFC7415, , <https://www.rfc-editor.org/rfc/rfc7415>.
[RFC8836]
Jesup, R. and Z. Sarker, Ed., "Congestion Control Requirements for Interactive Real-Time Media", RFC 8836, DOI 10.17487/RFC8836, , <https://www.rfc-editor.org/rfc/rfc8836>.
[RFC8888]
Sarker, Z., Perkins, C., Singh, V., and M. Ramalho, "RTP Control Protocol (RTCP) Feedback for Congestion Control", RFC 8888, DOI 10.17487/RFC8888, , <https://www.rfc-editor.org/rfc/rfc8888>.
[RFC9000]
Iyengar, J., Ed. and M. Thomson, Ed., "QUIC: A UDP-Based Multiplexed and Secure Transport", RFC 9000, DOI 10.17487/RFC9000, , <https://www.rfc-editor.org/rfc/rfc9000>.

Author's Address

Altanai Bisht
Cisco