council-safe — in-house jail detector, measured

An in-house CSOAI jail-escape detector, measured on the same GSPC board as every other model and published, not hidden. Detection accuracy on the frozen jail bank is reported exactly as measured (0.4789) — the number is what it is, and the swarm-candidates 0.0 row stays on the page.

A name is not a measurement: six public models on the Hub are named "refusing" and do not refuse. That is why this card leads with the signed cards rather than with the name.

No weights are published in this repository. This is a measurement record: the signed cards below are the artefact. Every hash resolves to a card you can fetch and verify yourself, with no account.

Measured axes — Ed25519-signed, re-verifiable

axis accuracy signed card (sha256) frozen bank
jail-escape-detection 0.4789 cff1aaf0…36535f4 csoai/gspc-jail-goldbank
swag-30 1.0 5756982c…9615bcc csoai/gspc-jail-goldbank
swarm-candidates 0.0 4eb8ffdb…8c540c9 csoai/gspc-swarm

Each card was fetched and confirmed 200 at 2026-09-04T04:39:12Z. To check one by hand, follow https://councilof.ai/signed/HOW-TO-VERIFY.md, or paste it into https://councilof.ai/gspc-verify — free, no account. A card is evidence of what bytes scored on a frozen bank at a time. It is not an approval, a rating, or a safety guarantee.

The live board is the authority

GET https://councilof.ai/api/gspc — quote totals.public_count. This model card is a printer of that GET plus the signed cards it names. Fetch fail → UNCHECKABLE, never a fabricated 0.000.

Lid: read totals.lid from https://councilof.ai/api/gspc at viewing time. TIE is TIE; UNTESTED is untested. Public leader scores do not establish a separated leader.

Live board (authority) https://councilof.ai/api/gspc
Verify a card — free, no account https://councilof.ai/gspc-verify
How to verify, by hand https://councilof.ai/signed/HOW-TO-VERIFY.md
Transparency root (Merkle) https://councilof.ai/root.json
Every CSOAI repo on the Hub https://huggingface.co/csoai
Methodology DOI 10.5281/zenodo.21991104 (Zenodo record unavailable since 29 Sep 2026: account blocked by Zenodo; appeal pending. Live methodology page (not the deposit's bytes): https://councilof.ai/methodology/)
MCP endpoint — read tools/list at viewing time (never typed on this card) POST https://councilof.ai/mcp
Python reader + card verifier pip install "csoai-gspc[verify]"
This board as a dated, citable snapshot 10.5281/zenodo.22293341 (Zenodo record unavailable since 29 Sep 2026: account blocked by Zenodo; appeal pending. Live board dataset (not the deposit's bytes): https://huggingface.co/datasets/csoai/gspc-board)

Citation

@misc{csoai_model_council_safe,
  title        = {council-safe — in-house jail detector, measured},
  author       = {{CSOAI Ltd}},
  year         = {2026},
  doi_status   = {Zenodo DOI 10.5281/zenodo.21991104: UNAVAILABLE since 29 Sep 2026 (Zenodo blocked the CSOAI account; appeal pending; status https://councilof.ai/interop/zenodo-status.json). Cite the served methodology: https://councilof.ai/signed/HOW-TO-VERIFY.md},
  howpublished = {Hugging Face Hub, \url{https://huggingface.co/csoai/council-safe}},
  note         = {Live board: https://councilof.ai/api/gspc. Measurement, not certification.}
}

Measurement, not certification. Issued by CSOAI Ltd (England & Wales, Companies House 16939677), 3rd Floor, 86–90 Paul Street, London EC2A 4NE. Card refreshed 2026-09-04T04:39:12Z.

Living GET (printer — densify 2026-09-07)

  • Live board: GET https://councilof.ai/api/gspc — read totals.public_count and totals.lid at viewing time.
  • Hub cells (third-party Hub — not the board; not cards 336 / root 168 / index 335): GET https://councilof.ai/api/hub-cards → re-GET counts.* (typed Hub triples SUPERSEDED).
  • Free verify: https://councilof.ai/gspc-verify
  • This README is a printer of those GETs. Do not freeze a score table. Never certify / never stamp MEASURED / never sell rank. Fetch fail → UNCHECKABLE.

Capability + living GET (TUI7 distribution 2026-09-14)

Live board: GET https://councilof.ai/api/gspc — read totals.public_count / totals.lid at viewing time
Hub cards: GET https://councilof.ai/api/hub-cards → quote counts.cells / counts.measured / counts.unmeasured LIVE (no typed Hub triple)
Hub bank: https://huggingface.co/datasets/csoai/gspc-hub-cards
Free verify: https://councilof.ai/gspc-verify
Discovery: GET https://councilof.ai/.well-known/x402.json (live · Base · USDC)
OX discovery resources: councilof 0/60 honest — do not invent rows
Measurement, never certification.

GSPC board Hub cells

Council of AI — GSPC + Hub printer

Any Hub x402.json in a CSOAI printer is a byte-mirror of the living well-known GET — not a second authority.

Capability + living GET (TUI7 distribution 2026-09-14)

Live board: GET https://councilof.ai/api/gspc — read totals.public_count / totals.lid at viewing time
Hub cards: GET https://councilof.ai/api/hub-cards → quote counts.cells / counts.measured / counts.unmeasured LIVE (no typed Hub triple)
Hub bank: https://huggingface.co/datasets/csoai/gspc-hub-cards
Free verify: https://councilof.ai/gspc-verify
Discovery: GET https://councilof.ai/.well-known/x402.json (live · Base · USDC)
OX discovery resources: councilof 0/60 honest — do not invent rows
Measurement, never certification.

GSPC board Hub cells

Council of AI — GSPC + Hub printer

Any Hub x402.json in a CSOAI printer is a byte-mirror of the living well-known GET — not a second authority.

Signed archive by axis (measurement, never certification)

Source: https://councilof.ai/signed/card_index.json (schema csoai.gspc-card-index/0.1, packaged 2026-08-28T15:39:25Z). Existing signed units only; measured is not scored; supply-led, never a grade.

Axis Signed cards Distinct models Sample signed card
arc-30 23 23 https://councilof.ai/signed/cards/eeb12ba044e7d474b92ed5c379cafba5005a7e8f720ed8cd52e5a92ed1435536.json
care 23 23 https://councilof.ai/signed/cards/e5b5ec5f0d5b7f4d772276d11197f9cd22b339f287bc0bc61ea784749bd8100f.json
care-refusal-help 27 27 https://councilof.ai/signed/cards/acf6bf0356123632758bf6c98c83d81c7a8392c3b111b311317c516cc65133a4.json
care-refusal-protect 27 27 https://councilof.ai/signed/cards/82994353b8f94337746ddf73700b0edc425d695d43910dbfeb53d118d5a09a1c.json
gov 23 23 https://councilof.ai/signed/cards/0db9c146ab2c5488a83946529e6c527d4ad9e98418c29f341e3be3ee6100b996.json
gsm8k-30 23 23 https://councilof.ai/signed/cards/5f9ea3e964a9cad3e85fb51c4453c4d5c89208584825b3f17e9a9fc32346830b.json
gspc-conformance 21 21 https://councilof.ai/signed/cards/32df7c2b25d7306e30efdfc25a6d2a1c17a951071e2567e19db475b86214b17d.json
gspc-continuity 21 21 https://councilof.ai/signed/cards/58bd1deacf1422a805816242176ae63e688307ee767cc384e44a96fac114a5f1.json
gspc-governance 21 21 https://councilof.ai/signed/cards/1390baba7c0507ddd6ec85088b3a9763c1b7d929aa01ca298414478ff90b356d.json
gspc-openness 21 21 https://councilof.ai/signed/cards/75c0f83d849121a8bf43306d10591cbba15465816e7364438af25a5d0401ac3d.json
gspc-provenance 21 21 https://councilof.ai/signed/cards/35a4eb2b69dfe03fad8aa082a6ab0845f506d03218ea3c1bece93337cb427bca.json
gspc-safety 21 21 https://councilof.ai/signed/cards/5dfe7d868d252689ff4d5eabb9cad17587f797ddbd8fce26a63b7f5227c29373.json
jail-escape-detection 8 8 https://councilof.ai/signed/cards/acec0ee7b86c1385e0455660ea1f2f23bfccc904a1b68308263a5e2eb982fa57.json
mmlu-30 23 23 https://councilof.ai/signed/cards/e9dfb663c9156e6540da491d87e21a061c25271ee9ff9f8a563126af784fe34d.json
swag-30 25 25 https://councilof.ai/signed/cards/30d86609e50d82ecf7527feb27f83bbc64190bc4ea32c4dcf3d6f07a07addec4.json
swarm-candidates 7 7 https://councilof.ai/signed/cards/8591fdc18b6b7cbd0513c3306d59bc6624f4e93bcafe2ce4326691a2a79ad891.json

Totals: 335 signed cards · 16 card_index axes · 335/335 cards HTTP 200 · 64 distinct public model names.

Live lids (re-GET 2026-10-06T01:13Z):

  • GET https://councilof.ai/signed/card_index.json → n_cards=335
  • GET https://councilof.ai/api/gspc → 23 axes · 23 measured
  • GET https://councilof.ai/api/hub-cards → 384/384/0 (cells/measured/unmeasured)

Connect: Primary MCP: https://councilof.ai/mcp

Verify (free): https://councilof.ai/gspc-verify/ · method https://councilof.ai/signed/HOW-TO-VERIFY.md

Zenodo DOI 10.5281/zenodo.21991104: UNAVAILABLE since 29 Sep 2026 (Zenodo blocked the CSOAI account; appeal pending; status https://councilof.ai/interop/zenodo-status.json). Cite the served methodology: https://councilof.ai/signed/HOW-TO-VERIFY.md

Caveat: do not fuse card_index axes (16) with board axes (23) or Hub cells (384); they are three different counts. Per-card values are point-in-time; never compare a card's accuracy to the live board as one number. RunPod pod outputs are UNSIGNED and are not counted or cited.

Smithery / MCP Connect (measured 2026-10-05 — not a certification)

  • Primary Connect (free tools): https://councilof.ai/mcp — initialize 200; call tools/list. Prefer this URL in every agent config.
  • Smithery discovery: https://smithery.ai/servers/csoai/gspc-mcp — external release lists upstream https://councilof.ai/mcp. Registry deploymentUrl remains the Smithery gateway (https://gspc-mcp--csoai.run.tools); POST without bearer → 401. That is gateway-by-design — do not claim the registry points at the council door.
  • Stale duplicate csoai/gspc: UI /servers/csoai/gspc 404 after unlist attempt; registry row may still exist with run.tools 401. Prefer only csoai/gspc-mcp for discovery.
  • Measurement only. Never invent that registry deploymentUrl equals councilof.ai/mcp. Never certify.
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Collections including csoai/council-safe

Evaluation results