Trust · AI provider disclosures
Where can our information go, and what is known about it?
Two features can send content to an outside AI service when a deployment is configured for it. For each, the register says what is known, and just as plainly what is not.
No certification, audit or independent assessment is held. Each statement below shows how sure the project is of it.
What exists, and how sure we are
- Reviewed in codeAI drafting of procedures starts from a recording's transcript and still frames, and sends them to a hosted language model when the deployment is configured for it.Recorded in the AI disclosure register.
- Tested locallyAI output lands only as a draft: a person approves it before it is published.Recorded in the security assurance record.
- Not establishedWhether the provider may use submitted content to train its models has not been established, so no statement about it is made.Recorded in the AI disclosure register.
- Not establishedHow long the provider keeps submitted content has not been established.Recorded in the AI disclosure register.
- Reviewed in codeDrafting uses global inference by default; regional inference has not been assessed, so content may be processed outside the organisation's region.Recorded in the AI disclosure register.
- Reviewed in codeSpeech transcription uses a separate provider whose processing location is recorded as unknown.Recorded in the AI disclosure register.
- Reviewed in codeA third path, governed AI answers, is not connected to any provider in production code.Recorded in the AI disclosure register.
- Not establishedWhich provider each AI feature calls, and what the agreements and account settings say, has not been verified for a pilot.Recorded in the pilot evidence pack.
See it in an example
Simulation. A scripted example on fictional data. The wording is scripted and no model is called; the figures are computed from what you decide, and a person’s approval covers exactly what you decided.
Decide what an AI provider may receive for drafting
- What happened
- Tidewell Foods (fictional) is considering switching on AI drafting of procedures and wants to decide which kinds of information may be sent.
- The question
- Which kinds of information may be sent to the drafting provider, and which must stay inside the platform? Asked by: Data protection lead at a customer (fictional).
Decide what an AI provider may receive for draftingChoose what may be sent. Restricted kinds cannot be chosen; a classification the analysis is unsure of must be confirmed by a person first.Open
Choose what may be sent. Restricted kinds cannot be chosen; a classification the analysis is unsure of must be confirmed by a person first.
Choose what the platform may read. Restricted material cannot be put in scope here, and a classification the analysis is unsure of must be verified by a person.
Read it? Source Holds Sensitivity What the analysis thinks it is Procedure text
Procedure library1,400 records internal Procedures clear Transcripts of recordings
Recordings620 records internal Spoken text check Recordings and still frames
Media storage340 records internal Frames from recordings check Names and email addresses
Directory2,100 records restricted People clear Workforce records
HR export1,900 records restricted Employment records clear Billing details
Billing410 records restricted Invoices clear Audit records
Evidence ledger88,000 records restricted Audit entries clear - Sources in scope
- 0 of 7
- records in scope
- 0
- Restricted, kept out
- 4 sources
- To verify
- 0
0 of 7 kinds of information may be sent, 0 records; 4 restricted kind(s) stay inside the platform.
The register lists three kinds of information for drafting: procedure text, transcripts and frames. Names, workforce records, billing details and audit records are not among them [S1]. 0 chosen kind(s) still need a person to confirm what they contain.
Even so, nothing is claimed about what the provider does with it: whether it may train on content, and how long it keeps it, have not been established [S2].
Relies on: S1 · The platform's AI disclosure register: what drafting receivesS2 · The same register: what has not been established
The customer's data protection lead decides. Approving records the customer's decision about which kinds of information may be sent. It does not switch drafting on, and an organisation-level off switch is not available today.
Not ready. Choose at least one source to read first.
Not done here. It makes no statement about training, retention or processing location, because none has been established, and it cannot switch an AI feature on or off.
What a reviewer can have, and what not yet
- We can hand overThe AI disclosure register for each feature, with what is known, what is the provider's own statement, and what is not established.
- We cannot yetProvider agreements and account settings verified for a pilot, a statement on training or retention, and a per-organisation off switch.
The capabilities behind this
- Govern the AI your organisation usesTeams adopt AI tools faster than anyone records them, who approved them or what they may touch.
- AI Governance AssuranceKeep an inventory of the customer's AI systems and agents: model and provider dependencies, risk assessments, use cases, approvals, policy obligations, human oversight…
The other nine
- Customer-controlled accessAccess by organisation, role and location, decided on the server, and what stops when it is withdrawn.
- Read-only database connectionsWhat a read-only connection proves before every use, what it cannot do, and what a customer's administrator should check first.
- Uploaded dataLimits, what is read and what is not, and the checks that are not done.
- Encryption and access controlsWhat protects sign-in, sessions and secrets, and the controls that only a live inspection can settle.
- Audit evidenceA tamper-evident decision record, what it proves, and what it cannot.
- Human approvalsWhat an assistant may draft or propose, what it refuses, and how a person's decision is recorded.
- Data retention and deletionWhat can be deleted now, what must wait, what is kept by design, and what is not finished.
- Customer-hosted and dedicated deploymentUpload, read-only connection, natural-language questions, a customer-hosted connector and a dedicated environment: what exists.
- Pilot security requirementsThe twenty-item gate, what it counts, and what is open.