Skip to main content

Controls, stated plainly

Clinical decisions stay human.Staff stay in control.

How Ask Brigid supports GDPR accountability and Irish health-data security, including the evidence still outstanding.

Built in IrelandEU data hostingGDPR accountability programme
Last updated: 27 October 2026

GDPR & Irish Data Protection Programme

Our documented privacy programme is designed around the Irish Data Protection Act 2018 and GDPR. The product includes workflows that help controllers respond to access, rectification, erasure, portability and objection requests. Compliance still depends on each clinic selecting a lawful basis, configuring the service appropriately, training staff, and completing its own accountability duties.

Patient preference records support workflow choices but do not themselves establish a clinic lawful basis
Article 28 Data Processing Agreement published for execution with each clinic customer
Article 30 Records of Processing Activities (ROPA): version 3.6, approved by the accountable director on 29 September 2026
Incident-response process supports the controller’s Article 33 assessment and 72-hour deadline

Where records live: AWS Dublin (eu-west-1)

The managed Postgres and object-storage service (Supabase) runs its EU project on AWS infrastructure in Dublin (eu-west-1), so a patient record created in Ask Brigid is written and stored on AWS Dublin from the first save. Some communications, monitoring and mapping paths use US processors (for example email delivery and map display); the Sub-Processor Register names each route, and several of them — including telehealth recording, SMS to patients and WhatsApp — are switched off, and several that are on still have their transfer instrument not yet captured.

Primary database and object storage: AWS Dublin (eu-west-1), via Supabase
The database is backed up daily and eight daily backups are kept, in the EU. A restore of a daily backup was tested on 1 October 2026 and matched
Point-in-time recovery is not on, so up to about 24 hours of data could be lost in a restore. Stored files are copied every 15 minutes to backup storage in the same project
Routes that use a non-EU processor are named in the Sub-Processor Register with their own transfer evidence
Availability commitments apply only where stated in an executed order form or service level agreement
Security controls are mapped against relevant HSE guidance; no HSE or other regional hosting certification is claimed

International transfers and AI processing locations

Primary patient records are hosted in AWS Dublin (eu-west-1). Each communications, monitoring, voice, AI and telehealth route has its own location and transfer analysis. Google’s Gemini models on Google Cloud Vertex AI, in Google’s EU multi-region, are used for drafting notes and letters the clinician reviews and signs, and for the administrative functions listed under EU AI Act readiness below; they are not approved for clinical decision-making.

Google’s Gemini models on Google Cloud Vertex AI, in Google’s EU multi-region, serve note and letter drafting and the administrative functions — pre-consultation chart summary, document-filing scan, template field detection, the patient-import identity read, mailbox sorting, emailed-document filing, the onboarding assistant and the Ask Brigid assistant. The pre-consultation chart summary may fall back to Frankfurt (europe-west3). Embeddings are processed in europe-west4 (the Netherlands). Production purpose, route and log evidence are internal records
Ask Brigid Live, the voice mode of the assistant, is switched on for staff. Before the microphone opens the user is told it is an AI and acknowledges how it works. Speech goes to Google’s Gemini Live model on Google Cloud Vertex AI in the EU, and Live uses administrative tools only. Reading patient records by voice is switched off for every practice
The former Google AI Studio / US-audio Live route is disabled and is not an opt-in fallback
Speech-to-text runs on Google’s Gemini transcription model on Google Cloud Vertex AI in the EU, with Google Cloud Speech-to-Text on an EU endpoint as the fallback; there is no ElevenLabs or US medical-model route. Original consultation audio saved through the standard route is deleted within 24 hours after the clinician accepts the note, and within seven days at the most. Recordings saved before 8 October 2026 or outside the standard route are being brought under that schedule; until then some may be kept longer
Video visits use Daily, Co. (United States). The call server is pinned to Frankfurt; account, room and usage metadata are processed in the United States. Nothing is recorded, transcribed or summarised, and recording of video visits is switched off
Web delivery and DNS for askbrigid.com and the Ask Brigid and MyBrigid web apps are provided by Cloudflare, Inc. (United States, global network), which replaced Vercel on 25 October 2026. It handles each request’s network details, such as IP address and page address. Clinical data passes between the browser and the Supabase database in Ireland and is not intended to be stored by Cloudflare. See the Sub-Processor Register
Controller-enabled vendor routes remain disabled until the applicable agreement, region, retention and access evidence is complete

Security controls and evidence

Provider-managed encryption at rest and transport protection are part of the intended control set. Production configuration, cipher and access evidence is still outstanding; this page does not turn an unevidenced setting into a guarantee.

Provider encryption and transport settings are checked against production evidence
Defined platform-security and clinical-access events are logged. Their retention period is drafted and awaiting approval; until it is approved and running, these logs are not deleted on a schedule
Staff sign in with a password, or with a one-time code or link sent by email and no password. Clinic staff and platform admins are asked again at least every 30 days. This has been in force since 1 October 2026. Authenticator-app verification is also available. Once an authenticator app is on, the email step does not replace it. One App Store review account is exempt. Privileged access was last reviewed on 31 August 2026
RLS is required on in-scope patient-data tables; database security-lint findings must be closed or formally accepted, and that evidence is still outstanding

Irish healthcare context

The product is designed for Irish private-practice workflows. Relevant HSE, professional confidentiality and record-keeping guidance is considered during the control review, but no HSE, Medical Council, CORU, Dental Council, NMBI, PSI, PCRS or HealthLink approval or certification is claimed.

Professional and HSE guidance is mapped in the working accountability pack and remains subject to review
No regulator is represented as having accepted the audit-log format
HealthLink, PCRS and other external routes remain gated until contract and configuration evidence is complete
Privacy, confidentiality and record-retention controls must follow applicable law, professional guidance and each controller’s documented schedule

EU AI Act readiness

The intended purpose is clinic administration, including drafting notes and letters that the clinician reads and signs. AI that diagnoses, triages or recommends, and patient-facing Generative AI, are disabled. The system inventory, role analysis and risk classification are maintained as internal records; the administrative routes in service on 28 September 2026 each have a written intended purpose declared by the accountable director (25 August to 28 September 2026), and the functions added since are being added to that written assessment. EU AI Act Article 50 transparency has applied since 2 August 2026, and Article 4 (AI literacy) applies to us and to each clinic. Ask Brigid’s AI functions are not intended for any use listed in Annex III of the Act.

Patient-facing Generative AI is disabled: the MyBrigid app has no AI chat or AI-written reply
AI runs only on administrative functions: note and letter drafting; tidying of dictated text for the clinician to review; pre-consultation chart summary; document-filing scan; template field detection (reading a blank form to find its fill-in fields); the patient-import identity read (filing an imported document to the right chart); mailbox sorting (labelling each email and putting it in a queue order for how soon staff should open it — an order for the mailbox, not a statement about a patient); emailed-document filing; the onboarding assistant; the Ask Brigid assistant, including its voice mode and its memory of staff preferences and open work; suggestions of administrative jobs for staff to approve, where a practice switches them on; and lookup of SNOMED and ICD codes and the US NPI registry on request. These functions remain subject to approved staff instructions and competence measures
Medical-literature search and drug-interaction and drug-label lookups are switched off
No AI output may diagnose, triage, prescribe or recommend care, and a draft is not part of the clinical record until a clinician signs it
A patient preference or terms acknowledgement cannot override the server-side purpose boundary

EU AI Act Article 50: what staff are told

Article 50 transparency duties have applied since 2 August 2026. Staff are told when they are talking to an AI system. AI-drafted notes, letters, summaries and email drafts carry a visible AI label where a person reads them before acting, and are marked as AI-generated in the stored record. Some parts are not complete, and they are listed below rather than left out. No patient talks to an AI system in Ask Brigid or MyBrigid.

The in-product assistant and the setup assistant identify themselves as AI, and Ask Brigid Live voice shows an AI disclosure before the microphone opens (Article 50(1))
AI-drafted notes, letters, summaries and email drafts are labelled as AI-generated where staff read them, and the stored record carries an AI-generated marker (Article 50(2) asks providers to mark AI output in a machine-readable way)
Not yet complete: AI-raised tasks are labelled when a task is opened but not in task lists; the category and priority given to an email are not labelled; and machine-readable marking of exported documents and of emails the assistant sends is being added. For AI systems placed on the market before 2 August 2026 the machine-readable marking duty applies from 2 December 2026
Patient-facing Generative AI and AI that diagnoses, triages or recommends remain disabled, so no patient talks to an AI system
The Ask Brigid assistant can draft a note or letter from the consultation, marked as AI-drafted, for the clinician to review and sign

EU Medical Devices Regulation (MDR)

DJG Media Limited, CRO No. 762838 does not claim a CE mark for the current product. The intended purpose is clinic administration. Ask Brigid can draft a note or letter from what was said in the consultation, for the clinician to review and sign; Generative AI that interprets, diagnoses, triages or recommends, and patient-facing Generative AI, are disabled. The manufacturer’s documented software-qualification determination, recorded on 10 September 2026, is that the intended purpose is administrative. It is the manufacturer’s position, not an HPRA decision. The determination was made before the note assistant (declared on 28 September 2026) and the later functions existed; they are being assessed under MDR Article 2(1) and MDCG 2019-11, and the 10 September determination does not yet cover them. This page is not regulatory approval.

Ask Brigid does not diagnose, treat, triage, prescribe, assign clinical codes, or recommend care
Clinicians author and release clinical notes, letters, referrals and prescriptions
No CE mark is held or claimed for the current product
See the Software qualification statement for the manufacturer’s determination and its limits

Certifications & Audits

We claim only what is evidenced today. We run a documented GDPR accountability programme. GDPR is not a product certification, and we do not claim that the product or the company is “GDPR compliant”. Some evidence and legal sign-off are still outstanding. The Aikido feed is currently disabled and the latest retained snapshot still contains unresolved findings, so no continuous-scanning claim is made. We do not currently hold SOC 2, ISO 27001, or HIPAA certifications.

GDPR — documented accountability programme; some evidence and legal sign-off are still outstanding
Aikido feed currently disabled; the retained findings and endpoint-coverage gap must be closed and fresh evidence captured
No EU MDR CE mark is held or claimed; the manufacturer’s determination of 10 September 2026 is the qualification record
No independent penetration test has been done, and no SOC 2, ISO 27001 or HIPAA certification is held or claimed