Trust & Security
Parenza holds a record of a child's health and routine. This page says what we do with it, control by control — and it says plainly what we have not finished.
Last updated 23 August 2026
The honest version
Parenza is not SOC 2 or ISO 27001 certified, has not completed a Data Protection Impact Assessment, and is not registered with the ICO yet. Formal work is under way and dated below. What exists today is the part a certificate is meant to prove — the controls in this page, each with its real status, built by someone who has been through these audits from the inside. We will not display a badge we do not hold.
What we hold, and why
- Routine entries you log — feeds, sleep intervals, nappies, pumping, growth measurements, and health events such as medication or vaccinations, each with the time it happened. This is what makes an answer specific to your child.
- Family context you write — profile, household facts, the principles you've agreed on as parents, and a decision log. This is what the panel reads before it answers.
- Questions and answers — so the reasoning behind past advice stays available to you.
- Account basics — your name, email address, and sign-in identity.
No advertising identifiers. No data brokers. No third-party analytics in the product, and no tracking script on this website. We do not sell your data and we do not use it to target advertising — those are not lines we are keeping open for later.
Where it runs
Parenza is an independent product operated on SHIP platform infrastructure — the same API, application and agent runtime, running on European servers (Hetzner, Finland). Your family's data sits in its own tenant, scoped at the query layer, and is not pooled with other families' records. Traffic runs over TLS; stored third-party credentials are encrypted at the application layer with keys held outside source control.
Where we stand, framework by framework
“Designed to” means covered by architecture and process, and independently verifiable in an audit we have not commissioned yet.
UK GDPR & Data Protection Act 2018 In force today The one set of obligations that already applies to us, and the backbone of the promise that this record is yours. Rights are meant to be product features, not an inbox — some already are, and the rest are dated. Show control detail + Hide control detail −
- Lawful basis & minimisation (Art. 5/6) In place
We collect what you choose to log and what the panel needs to answer you. No advertising identifiers, no data brokers, no third-party analytics in the product.
- Right of access & portability (Art. 15/20) In progress
Export is produced for you on request today; a self-serve machine-readable export is in development.
- Rectification (Art. 16) In place
Every routine entry and knowledge document is editable in the app.
- Erasure (Art. 17) In progress
Honoured on request within 30 days today, actioned by a person; automated cascade deletion is being built before general access.
- Records of processing (Art. 30) In progress
Being written as part of the DPIA, alongside the retention schedule.
- DPIA for high-risk processing (Art. 35) In progress
Required because this is children's health data. In progress; completion is a gate on onboarding beyond the design-partner cohort.
- Breach notification (Art. 33/34) In progress
The 72-hour ICO clock and the family-notification path are written into the incident plan; the plan has not been exercised yet.
- ICO registration Planned
Before general availability, with the DPIA.
ICO Children's Code (Age Appropriate Design) Designed to · self-assessment in progress Parenza holds data about a child who cannot consent, entered by a parent who can. The Code's standards are the sharpest test of this product's design, so they get their own row. Show control detail + Hide control detail −
- Best interests of the child In place
Specialists are built to refuse age-inappropriate advice and to hand you to a clinician rather than answer past their remit.
- Data minimisation In place
Nothing about the child is collected automatically. Every entry is one a parent chose to make.
- No profiling for commercial purposes In place
No advertising, no selling of data, no behavioural profiling. Reasoning happens only in answer to a question you asked.
- Detriment & nudge techniques In place
No streaks, no guilt mechanics, no engagement-maximising notifications. Missing a day of logging costs you nothing.
- Formal Code self-assessment In progress
Documented alongside the DPIA.
SOC 2 — Trust Services Criteria Designed to · audit when there is a business to audit A Type II report needs months of operating evidence from a real company. That evidence trail is being built now; the architecture it will be tested against already exists. Show control detail + Hide control detail −
- Logical access (CC6) In place
Federated sign-in with no password database, per-family scoping enforced at the query layer, least-privilege service credentials.
- Change management (CC8) In place
Contract-first API, generated clients, CI gates on lint, types and tests, infrastructure as code — no hand-edits in production.
- System operations & monitoring (CC7) In progress
Health checks and an audit trail for AI tool calls are live; centralised alerting across the whole surface is not finished.
- Risk & vendor management (CC3/CC9) In progress
Subprocessor register is published below; a formal risk register is being maintained as part of the DPIA work.
- Confidentiality (C1) In place
Per-family isolation, credentials encrypted at the application layer, secrets outside source control with separate keys per environment.
- Availability (A1) In progress
Nightly automated backups run to separate storage; scheduled restore testing with published RPO/RTO targets is the next resilience item.
ISO/IEC 27001:2022 Designed to · certification with productization An ISMS scaled honestly to a company of this size: the Annex A technological controls implemented in code rather than described in a binder, with the organisational half still being written. Show control detail + Hide control detail −
- Organizational & people controls (A.5/A.6) In progress
Policy set is being adapted from the platform's existing compliance documentation to Parenza's own scope.
- Identity & access (A.5.15–A.5.18) In place
Federated identity, scoped tokens, per-client authorization for AI assistants, revocable individually.
- Cryptography (A.8.24) In place
TLS on every public endpoint; stored third-party credentials encrypted at the application layer with keys held outside the repository.
- Logging & monitoring (A.8.15/A.8.16) In progress
Every AI tool call against your family's data is logged with actor, tool and timestamp. Coverage of ordinary app actions is being extended to match.
- Secure development (A.8.25–A.8.28) In place
Typed, parameterized persistence; validation against a single API contract; review and CI gates before merge.
- Vulnerability management (A.8.8) In progress
Dependency auditing runs in the workspace; automated scanning and update policy across every repository is being rolled out.
Controls, category by category
Access control
- Passwordless sign-in In place
Google OAuth. Parenza never sees or stores a password, so there is no password database to breach.
- Multi-factor authentication In place
Delegated to your Google account — if MFA is on there, it protects Parenza. We do not offer a bypass.
- Per-family data isolation In place
Every read and write is scoped to your family's tenant at the query layer. There is no cross-family access path in the API.
- Per-client AI authorization In place
Each assistant you connect authorizes separately over OAuth and can be revoked on its own.
- Rate limiting on unauthenticated surfaces In place
The public OAuth endpoints are rate-limited per IP.
- Named, separated production access In progress
Operational access is limited to the founder today. Role separation is being formalised as the team grows.
Data security
- Encryption in transit In place
TLS on every public endpoint, certificates issued and renewed automatically.
- Application-layer encryption of stored credentials In place
Third-party keys and tokens are encrypted before they are written, with the master key held outside source control and separate per environment.
- Automated backups In place
Nightly logical database backups, compressed and written to separate volume storage.
- Tested restores with published RPO/RTO In progress
Backups exist; a scheduled restore drill and published targets do not yet.
- Self-serve export In progress
Export is produced on request today; the one-click version is being built.
- Automated erasure & retention purge In progress
Deletion is honoured by a person on request; the automated cascade and a published retention schedule are in development.
Infrastructure
- EU hosting In place
Application and database run on European infrastructure (Hetzner, Finland).
- Infrastructure as code In place
Servers, networking and volumes are declared in Terraform with remote state — not clicked together in a console.
- Secrets outside source control In place
No credentials in the repository or in images; a pre-commit check warns on secret-shaped strings.
- Private binding for data services In place
Database and cache are bound to private addresses, not exposed publicly.
- Environment separation In progress
Separate keys and configuration per environment today; fully separate infrastructure stacks are planned.
Application security
- Contract-first validation In place
Every request is validated against a single API contract, and clients are generated from it rather than hand-written.
- Injection-safe persistence In place
Typed, parameterized queries only — no string-built SQL.
- Photo metadata stripped before upload In place
A child's photo added to the app is cropped and resized in your own browser before anything is sent. The re-encode keeps the pixels and discards the metadata, including the GPS coordinates a phone photo carries — for most families, their home address.
- Security headers & CSP In progress
Standard protective headers are sent; the Content Security Policy runs in report-only mode while auth flows are validated.
- Dependency auditing In progress
Audited on change today; scheduled automated scanning across every repository is being rolled out.
- External penetration test Planned
Commissioned before general availability. The summary will be published here.
AI & data governance
- No training on your data In place
Model providers are used under terms that do not permit training on customer data. Your family's record is never a training set.
- Consent-gated assistant access In place
An outside assistant can reach your data only through a connection you created and can revoke.
- Audit trail for AI tool calls In place
Every tool call an assistant makes against your family's data is recorded with the account, the tool and the time.
- Scoped responses In place
A request returns the records that answer it — connecting does not hand over a copy of your whole record.
- Stated limits on advice In place
Each specialist carries a written remit and escalation rules; medical, psychological and financial questions are routed to professionals rather than answered past competence.
Organizational
- Responsible disclosure In place
A published security.txt and a monitored address; reports go straight to the person who writes the code.
- Policy pack In progress
Adapted from the platform's existing compliance documentation to Parenza's scope. Available to design partners on request as sections are finished.
- Incident response plan In progress
Written, including the regulatory clock and the family-notification path. Not yet exercised.
- Subprocessor register In place
Published below and updated when it changes.
Connecting your own AI assistant (MCP)
This is the part worth reading slowly, because it is a mechanism most people haven't met before.
Parenza can be exposed as an MCP server — an address you paste into Claude, ChatGPT, Cursor or a similar assistant. Once you've added it and signed in to authorize the connection, that assistant can ask Parenza questions on your behalf: what this week's feeds looked like, what's in your decision log, what the feeding specialist said.
- Nothing leaves until you connect. The connection doesn't exist unless you create it, and each assistant authorizes separately, so revoking one doesn't affect the others.
- Data goes where the request comes from. When your assistant asks Parenza something, the answer travels to that assistant's provider — Anthropic if you're using Claude, OpenAI if you're using ChatGPT — and is then handled under their terms and your account settings with them, not ours. If your account there allows training on your conversations, that setting applies to what your assistant pulled from Parenza. Check it before you connect.
- Only what the question needs. Each request returns the specific records that answer it. Connecting does not hand over a copy of your whole family record.
- Every call is logged. Tool calls made against your family's data are recorded with the account, the tool and the time, so the connection is reviewable rather than opaque.
If you would rather keep everything inside Parenza, don't connect an assistant. The app works fully on its own.
Documents, on request
The policy pack behind this page. Documents marked on request exist today for the infrastructure Parenza runs on; the ones in draft are the Parenza-specific versions being written alongside the DPIA. Email security@parenza.ai to request any of them — an NDA may apply to some.
- Information Security Policy On request
- Access & Identity Policy On request
- Infrastructure & System Security Policy On request
- Backups, Logging & Recovery Policy On request
- Incident Response & Communication Plan On request
- Continuity & Resilience Plan On request
- Vendor & Third-Party Risk Policy On request
- Data Protection & Compliance Policy On request
- Detection & Monitoring Policy On request
- Data Protection Impact Assessment (children's data) In draft
- Record of Processing Activities (Art. 30) In draft
- Data Retention & Deletion Schedule In draft
- AI & Data Use Policy In draft
Subprocessors, in the open
Every vendor that touches data, what it sees, and where it sits.
| Vendor | Purpose | Data | Location |
|---|---|---|---|
| Hetzner Online | Hosting — application servers, database, backup volumes | All family data at rest | Finland (EU) |
| Google (OAuth) | Sign-in identity | Account identity — email address, account subject id | Global |
| Anthropic | Model inference for the specialist panel and briefs | The family context relevant to the question you asked | US |
| GitHub (Pages) | This marketing site only — no product data, no analytics script | IP address and request metadata of site visitors | Global (US entity) |
An AI assistant you connect yourself is not a subprocessor: it is your account with that provider, acting on your instruction. What it reads from Parenza is handled under your terms with them — which is why the MCP section spells this out rather than burying it here.
Public commitments, with dates
Gaps stated openly are promises, not weaknesses. These are the next milestones — hold us to them.
- Data Protection Impact Assessment completed Before general access
- ICO registration Before general access
- Self-serve export and automated account deletion Before general access
- Published retention schedule with automated purge Q4 2026
- Scheduled restore drills with published RPO/RTO Q4 2026
- External penetration test, summary published here Before general access
- SOC 2 Type II and ISO 27001 audits When there is a company to certify
Your data, your exit
The strongest security posture is not needing to trust us forever. Ask and we will export your family's record in a machine-readable form, or delete the account and its data — actioned within 30 days, by a person, until the self-serve versions ship. A record of your child's first year should never depend on whether a company survives.
Email hello@parenza.ai for either.
Found something? Tell us first
Responsible disclosure is welcome and goes straight to the engineer responsible for the code. Email security@parenza.ai with the details, or use security.txt. Act in good faith and within scope and we will not pursue legal action.
Security questions, answered
Why is there no SOC 2 or ISO 27001 badge on this page?
Because we don't hold one, and a badge we haven't earned would be the first lie we told you. The controls those audits examine are listed above with honest statuses. The founder has worked inside regulated businesses through SOC 2, ISO 27001 and DORA audits, which is why the architecture is built to that shape now — so certification is a milestone rather than a rebuild.
Who can see my family's data?
You, and anyone you invite into your family's account. Every query is scoped to your tenant at the data layer, so there is no cross-family access path in the API. Operationally, production access exists for the founder to run the service — used for operations, not for reading family records. An AI assistant sees data only through a connection you created and can revoke.
Do you train AI models on my child's data?
No. Model providers are used under terms that don't permit training on customer data, and we don't build models of our own. If you connect your own assistant, whatever it pulls from Parenza is then governed by your settings with that provider — worth checking before you connect.
What happens if Parenza shuts down?
You take the record with you. Export is produced on request today and self-serve export is being built before general access. A family's history of feeds, sleep and decisions should never be hostage to whether a company survives.
Is this a medical device?
No. Parenza is not a medical device, is not registered with the MHRA, and does not diagnose or prescribe. The specialists summarize published guidance and are built to escalate — to your GP, health visitor, NHS 111, or emergency services — rather than answer past their competence.
How do I report a vulnerability?
Email the security address below or use /.well-known/security.txt. Good-faith research is welcome: stay within scope, don't touch other families' data, and we won't pursue legal action. Reports go straight to the engineer responsible for the code — acknowledgement within 72 hours. There is no bug bounty yet.
What Parenza is not
Not a medical device, not a clinician, and not a regulated financial adviser. The specialists summarize published guidance — NHS, NICE, the Lullaby Trust, and other national guidance where it differs — and are built to hand you to a professional when a question needs one. Emergencies go to your local emergency services, not to an app.