AI assistants
Optional AI features: “Ask SimpleAct”, AI suggestion for findings, checking a DPA with AI and drafts for DPIAs and data breaches. The AI proposes, a person decides and saves.
The AI assistants take reading and drafting work off your hands: they answer questions about your data, explain findings, check a DPA and draft content for a DPIA or a data breach. All features are optional and only run when the AI assistant is switched on for your organisation. They only ever deliver suggestions: nothing is saved, notified or approved automatically. A person reviews every suggestion, takes it over into the form and saves it.
At a glance
- Included from
- Professional (AI assistants, beta)
- Where in the app
- Switch on: Settings → Workspace → “AI assistant (Claude, Anthropic)” (admin). The features themselves live in the respective modules.
- Output
- Answers with sources, explanations and drafts that you review and take over yourself
- Legal basis
- Art. 28, 33, 34, 35 GDPR
Switching on, daily limit and provider
- The AI assistant is switched off by default. An admin switches it on or off for the whole organisation in the settings; every change is recorded in the audit log.
- Cloud model (Claude by Anthropic): the content of the feature used is sent to Anthropic for processing. Anthropic is based in the USA, i.e. in a third country (Art. 44 et seq. GDPR); processing takes place under a data processing agreement. This is also stated right next to the switch in the settings.
- Independent of the switch: the in-app support chat (“Help”) sends every question to Anthropic to be answered, even when the AI assistant is switched off. It does not send data from your workspace; please do not enter personal data there.
- For the cloud model, each organisation has a daily limit of requests and a daily budget in euros. The settings show how much of it has been used today. Once either is reached, the features are unavailable until the next day.
- Self-hosted model: if a local AI model runs on your installation, no data leaves the infrastructure. The switch has no effect there, and the daily limit and budget do not apply. “Ask SimpleAct” is not available with a local model.
- Each feature also needs the permissions of the module it lives in, for example the edit permission for DPIAs or data breaches.
The AI notice on every feature
Every AI feature states before it starts what is read, for example: “An AI model reads the uploaded agreement; the text is sent to the AI provider for this. The result is a suggestion and may be wrong.” With a self-hosted model, the clause about the transfer is omitted. Which provider that is and where it is based is stated in the settings under “AI assistant”.
The features
Ask SimpleAct
A chat in the app header for questions about your own compliance data, such as “Where is a DPIA missing?” or “What is due in the next 30 days?”. The AI only reads, using the read-only tools of the MCP interface, run with your permissions: it never sees more than your own pages. Every answer names its sources (AI systems, processing activities, incidents, data breaches, data subject requests, AI intake requests). The conversation is not stored and is lost on reload. Visible to everyone with the permission to read the dashboard once the AI assistant is switched on.
AI suggestion for consistency findings
On a consistency finding, the AI explains in plain language what does not fit and recommends: change the entry, probably intended, or review manually. You take proposed values over into the form with “Adopt the AI’s suggestion” and save there. If the AI considers the finding intended, “Mark as intended with the AI’s draft reason” opens the dialog with the reason prefilled; you confirm it yourself. Errors and duties get no such draft, as those findings cannot be marked as intended.
Check DPA with AI
In the vendor register you upload the DPA as PDF or TXT (at most 5 MB). The AI rates each of the 15 mandatory contents under Art. 28(3) and (4) GDPR as covered, partly covered, not found or not assessed, and backs it with a verbatim quote. SimpleAct checks every quote against the document text; a quote that does not appear there is discarded. Buttons take detected details over into the form: contracting party, signing date, missing sub-processors, third-country transfer and ticking off the covered items. The document is not stored.
Draft with AI: DPIA
From the saved DPIA and the linked data you may read, the AI drafts the content under Art. 35(7)(a)–(d) GDPR: description of the processing, necessity and proportionality, risks with proposed likelihood and severity, and remedial measures, referring to existing TOMs where possible. You adopt or discard each part individually; if the field already holds text, the draft is appended rather than replacing it. The AI never sets the residual risk, the approval or the decision on a prior consultation (Art. 36); at most it gives a hint on the consultation.
Draft with AI: data breach
The AI drafts the content of the notification to the supervisory authority under Art. 33(3) GDPR (nature of the breach with categories and approximate numbers, likely consequences, measures), assesses whether a high risk under Art. 34(1) is likely and, if so, drafts the communication to the data subjects for you to copy. The app takes the contact of the data protection officer (point b) from the breach record or the DPO profile and computes the status of the 72-hour deadline itself; neither comes from the AI. The AI marks nothing as notified.
What the AI receives and what it does not
- Only what you may read yourself. Linked data without read permission is left out; the draft points this out.
- For the AI suggestion on a finding: rule, title, message, severity, legal references, the entries concerned with type and name, and the stored values of the fields concerned, nothing else.
- For a data breach, no contact data. Uploaded DPA documents are processed in memory only.
- Every tool call made by “Ask SimpleAct” is recorded in the audit log with the source “AI assistant (Ask SimpleAct)”.
Endpoints
Each route has a status route (GET) that reports whether the AI can answer right now and why not. If it is unavailable, there is no error but an empty result with a reason (switched off, daily limit, daily budget, unreachable, no usable result). At most 10 requests per minute per route and person or API key.
| Route | Permission | API key |
|---|---|---|
| POST /api/v1/assistant/chat | Read dashboard | No, browser session only (assistant.session_only) |
| POST /api/v1/consistency/findings/ai-suggestion | Only findings you see in full | Yes |
| POST /api/v1/vendor-register/{id}/dpa-check | Edit vendor register | Yes |
| POST /api/v1/dpia/{id}/ai-draft | Edit DPIA | Yes |
| POST /api/v1/data-breaches/{id}/ai-draft | Edit data breaches | Yes |
A suggestion does not replace a review
AI results may be wrong or incomplete. The DPA check is not a legal review, a DPIA draft is not the controller’s assessment, and an assessment of high risk is not a decision on the communication. Review every suggestion before you take it over.