Every action on the books, witnessed.
Four clinical AI systems governed at the boundary. Every inference and tool call is resolved against the signed charter and leaves a receipt — hash-chained, exportable, verifiable without Glacis.
Governed systems
charter-bound · receipts current| System | Risk tier | Charter | Active controls | Receipts today | Status |
|---|
Receipts
Each row is a signed decision record — not a log line. Select one and verify it yourself: the checks run in this browser, against the bytes on screen.
| Time | System | Control | Action | Receipt |
|---|
Why this standard is here to stay
The shift from self-attested logs to independently verifiable evidence is not a vendor thesis — it is written into instruments that bind deployers now. OVERT receipts are the operational answer.
| Authority | Instrument | What it requires | Effective | OVERT answer | Status |
|---|
Volatile specifics (dates, citations) summarized for briefing — VERIFY against source before client use.
One receipt schema, every framework
capture once · evidence everywhere| Evidence property of the receipt | ISO 42001 | NIST AI RMF | EU AI Act | HIPAA | TJC / CHAI | Insurer DDQ |
|---|
● satisfied by receipt fields ◐ partial / paired with charter artifacts — excerpt of the full mapping engine.
Stewardship
The standard does not belong to Glacis — that is the point. A record only counts as evidence when the party checking it doesn't have to trust the party that produced it.
Operationalize in two weeks
No model changes, no prompt changes, no cloud dependency. Glacis installs at the boundary the client already owns and starts writing the books on day one.
Install at the boundary
Unmodified LiteLLM proxy + callback for model traffic; MCP wedge for agent tool calls. On-prem.
litellm_settings:
callbacks:
- glacis.GovernanceLogger # receipts on every callCharter signed as code
The governing document becomes a signed policy: rules → named runtime controls → calibration, activated under the client's own key.
Receipts flowing
Baseline posture visible: every decision on the books, held-for-review queue live, drift measured against the attested baseline.
First evidence pack
Framework-mapped export with the standalone verifier — the client's auditor checks it offline, without trusting Glacis or the client.
Evidence packs
exports ship with the verifierISO/IEC 42001 readiness
EU AI Act Art. 12 extract
Insurer underwriting extract
Incident reconstruction
Run your advisory engagement on the instrument.
Your healthcare clients must move fast on AI without betting the institution. Glacis gives your engagement a measurable spine: a 30-minute assessment, a charter workshop, and a two-week evidence baseline — producing the artifacts their board, regulator, and insurer actually accept.