MITHDUIN← mithduin.com
Trust centre

Where the honest answer is “not yet,” this page says not yet.

Controls, documentation and answers for anyone doing diligence on Mithduin — a client, their counsel, or their family office. What the architecture enforces is stated as enforced; what is not yet in place is named, below, as not yet.

Get in touch

Compliance, honestly stated

Mithduin does not yet hold SOC 2, ISO/IEC 27001 or similar third-party certifications, and you will not find their badges here until an auditor has signed them. Practices are GDPR-aligned — data minimisation, export and deletion on request, named sub-processors — and the roadmap below is the commitment.

Roadmap.

SOC 2 readiness assessment; external penetration test with the executive summary published here; two-factor sign-in and passkeys for every account. Each item moves from this box to the controls list when it is real.

Controls

Infrastructure security

  • Row-level security on every table, enforced by Postgres per query.
  • TLS in transit; AES-256 at rest; private storage with five-minute signed URLs.
  • Functions run with the caller’s own credentials — no service key in any client path.
  • Secrets held in the platform’s secret store, never in source.

Product security

  • Read-only brokerage tokens; no trading permissions exist anywhere in the system.
  • SHA-256 recorded per uploaded document.
  • Verified releases: every deploy is checksum-gated so an altered file cannot ship.
  • Sign-up is invitation-oriented; every access and change is logged.

Data & privacy

  • Your record is never sold and never used to train models.
  • AI suggestions send structured field summaries only — never documents.
  • Export or deletion of the whole record on request.
  • Sub-processors: Supabase (database, auth, storage), Vercel (hosting), Anthropic (suggestion drafting), Interactive Brokers (read-only market data), Resend (transactional email).

Organisational

  • A deliberately small operation: fewer hands, fewer places for data to leak.
  • Oversight pairs AI agents in a hardened harness with trusted advisors across international tax (IRS, CRA, HMRC, EU regimes), structuring, trusts & estates, asset protection and finance.
  • Staff access to client records is role-based and visible to the client in the activity log.
  • Concerns are reviewed by the operating team and answered directly — see reporting below.

Resources

Mithduin overview — the one-pagerFeatures, what it computes, and security — printable.The security noteArchitecture and controls, stated plainly.Architecture & data-flow note — on requestDetailed write-up of tenancy, storage and the AI data path.

Frequently asked

What data encryption standards do you use?

TLS for every connection in transit and AES-256 at rest. Documents live in a private bucket with no public URLs and are retrieved through signed links that expire after five minutes. Each new upload is fingerprinted with SHA-256 so its integrity can be verified later.

Who has access to our data?

You, anyone you invite, and the advisory team that services your record — nothing else. The boundary is row-level security evaluated by the database on every query, not application code. Staff actions appear in the same activity log you see.

Do the AI models retain or learn from our data?

No. Suggestions are generated from deterministic rules over your record; the model only drafts the wording from a structured summary of relevant fields. Documents are never sent. The API is used under commercial terms that prohibit training on your data.

Do you conduct regular penetration testing?

Not yet — an external penetration test is on the near-term roadmap, and the executive summary will be published on this page when complete. We would rather tell you that plainly than imply otherwise.

Can we get our data out?

Yes. Export of the complete record — data and documents — or full deletion is available on request at any time.

Reporting a concern

If you believe you have found a vulnerability, or an adverse impact related to Mithduin’s AI features, write to office@mithduin.com. Reports go straight to the operating team, are acknowledged, and are remediated with priority over feature work.