Where the data goes — and who is on the hook for it.
This page covers what AMS does on your behalf. Your direct SaaS relationships — Anthropic, Airtable, Google, Microsoft, etc. — are between you and those vendors, the same way they are today for any tool you already use.
How AI-First handles your data
Three layers, three different places:
| Layer | Where it lives | Who controls it |
|---|---|---|
| Your business data | Your existing storage — Google Drive, OneDrive, Airtable, etc. | You |
| In-flight processing | Briefly transits to Anthropic when a Claude session asks a question | Anthropic (per their DPA with you) |
| AI-First service data | A small set of AMS-side services — see "What AMS holds" below | AMS (per the AMS DPA with you) |
What Claude actually does with your data
- Reads documents and records directly from your storage when asked.
- Sends only the relevant snippet to Anthropic for processing — never your full library.
- Returns the answer; nothing is indexed, embedded, or kept in a searchable database.
- Per-session — no persistent memory between sessions.
- Anthropic's API tier does not use inputs or outputs to train their models.
Your SaaS relationships — where the compliance sits
You already use SaaS tools today (Gmail, Asana, Xero, etc.) and accept their terms directly. AI-First adds Anthropic to that list. The compliance documents for each vendor live with that vendor:
| Vendor | Their compliance documents |
|---|---|
| Anthropic (Claude) | trust.anthropic.com, privacy.claude.com, DPA in Console → Settings → Legal |
| Airtable (your base) | airtable.com/security, airtable.com/trust |
| Google Workspace | cloud.google.com/security/compliance |
| Microsoft 365 | microsoft.com/trust-center |
For each, your compliance team can pull certifications (SOC 2, ISO 27001, etc.), the current DPA, and sub-processor lists directly from the vendor — same as they would for any tool you already use.
On US data transfer specifically
Every major cloud SaaS (including all the ones above) processes data in the US. They are GDPR-compliant on the basis of Standard Contractual Clauses included in their DPAs, the EU–US Data Privacy Framework where applicable, or both. This is normal practice — your lawyers will recognise the mechanisms.
What AMS holds and processes for you
This is the bit AMS is on the hook for. Everything else (above) is between you and the underlying vendor.
| Service | What it holds | Where it is hosted |
|---|---|---|
| AI-First Delivery (Airtable) | Interview transcripts, summaries, contact names & emails | Airtable (AMS workspace) |
| Interview landing pages | Interview record IDs, email allow-lists | Cloudflare Worker + KV |
| Interview invite sending | Recipient email addresses, in transit only | Resend (US) |
| Make scenarios (AMS-side) | Whatever the scenario processes — typically transient | Make (EU servers, Czech Republic) |
Retention
Personal data we hold (Airtable + Cloudflare): retained for the duration of our working relationship, plus 6 years afterwards — matching the UK statutory limitation period for contract claims. The basis is defence of legal claims and continuity of the ongoing client relationship (clients commonly re-engage and the historical context matters).
Vendor-managed logs (Resend, Make): per each vendor's standard retention — typically days to weeks for transient processing logs.
Earlier deletion or anonymisation is available on request, subject to any legal hold. After the retention window, personal data is either deleted or anonymised (identifiers stripped) while non-personal artefacts (workflows, architecture, patterns) are retained indefinitely for AMS reference.
For these: AMS acts as a processor under GDPR. An AMS DPA covers them — available on request. The DPA defines retention, deletion process, breach notification, sub-processors, and audit rights.
Sub-processors AMS uses
When I process your data on the services above, I use the following sub-processors. Each has its own DPA with me, and your compliance team can review their security posture directly on their trust centres:
| Sub-processor | What they process for AMS | Location | Their trust centre |
|---|---|---|---|
| Anthropic (Claude) | In-flight prompts during AI sessions | US | trust.anthropic.com |
| Airtable | AI-First Delivery base (transcripts, contacts) | US | airtable.com/trust |
| Cloudflare | Interview landing pages (record IDs, email allow-lists) | Global edge network | cloudflare.com/trust-hub |
| Resend | Invite email delivery (recipient addresses, briefly) | US | resend.com/legal |
| Make | AMS-side workflow automation (transient processing) | EU (Czech Republic) | make.com/en/security |
Any change to this list will be communicated to you before it takes effect.
For everything else (the bulk of where data lives — your Drive, your own Airtable base, your Gmail, Anthropic itself): your direct vendor relationships, your direct DPAs.
Confidential documents
AMS sets up a dedicated AI-First folder structure on your storage when we start working together. The AI only reads from that folder. Anything outside it — your other Drive files, your wider Airtable base, anything sensitive — is not visible to the AI.
The model is opt-in by location:
- Inside the AI-First folder: anything you put here can be read by the AI to help with the work.
- Outside the AI-First folder: the AI cannot access it. No risk, no configuration needed on your side.
When something new comes in — a draft contract, an IP filing, a confidential report — the decision is just "do I want the AI to help with this?" If yes, drop it in the folder. If no, leave it where it is. The boundary is the folder location, not a Claude setting that could be flipped.
For regulated sectors (financial services, healthcare, legal privilege), additional technical controls beyond folder configuration are usually required. Discuss with AMS upfront if that applies — the default AI-First setup is designed for unregulated and lightly-regulated SMEs.
Run your own compliance review
Before adopting any new SaaS — Anthropic, Airtable, and any other tools we connect to your operation — your team should run their own compliance review. This is your data and your business; that review is yours to own, not something you should outsource to me or any other implementer.
AMS sits in a different category: I am a consultancy and implementation partner, not a SaaS vendor. For the limited data I process on your behalf (see "What AMS holds" above), the AMS DPA covers what your team needs.
Below is where to find everything.
- Anthropic — trust.anthropic.com for certifications and sub-processors, privacy.claude.com for the DPA and Privacy Centre.
- Other SaaS — the trust / security page of each vendor you use, links above.
- AMS — the "what AMS holds" and sub-processor list are above; request the AMS DPA from jason@amsuk.co (issued per engagement).
If your compliance lead wants a 15-minute call to walk through specifics, book direct with jason@amsuk.co — happy to support the review, not replace it.