AI transparency
AI you can put your licence behind.
A pharmacy's name is its most valuable asset, and we treat it that way. Our AI does the work of a full growth team — but it proposes, you approve, and it runs only inside guardrails. Here is exactly how that works, and how it maps to the EU AI Act.
Last updated 2 July 2026
At a glance
- The AI proposes. You approve. It runs — nothing consequential happens without a human.
- Every AI-generated asset is labelled and carries a provenance record.
- No medical advice, ever — marketing and operations only.
- A claims firewall gives every outbound asset a versioned policy decision.
- You can see, stop, and override the AI at any point.
- Built to EU AI Act Article 50 transparency, effective 2 August 2026.
On this page
The principle
Supervised intelligence, not autopilot
We don’t sell “fully autonomous” — we sell your team, multiplied, and kept on a leash. Every workflow in Farmacy OS follows one shape:
01
Propose
The AI drafts the post, the ad, the follow-up, the campaign — grounded in your brand and data.
02
Approve
You review and one-tap approve. Consequential actions cannot skip this step.
03
Run
Only then does it publish, send, or spend — inside hard-coded limits.
Why this is the stronger claim
Human oversight
Human approval on every consequential action
Anything that touches the public, a customer, or money is gated. The AI cannot cross these lines alone:
| Action | Gate before it happens |
|---|---|
| Publish a post or ad | Claims-firewall pass + your approval |
| Send a WhatsApp / message | Consent + quiet-hours check + your approval |
| Change ad spend | Spend-limit check + reserve-then-settle + your approval |
| Message a customer segment | Opt-in verification + your approval |
| Draft, research, summarise | Runs freely — it's reversible and internal |
These aren’t polite suggestions to the model — they’re deterministic code gates that execute regardless of what the AI “wants” to do.
Provenance
AI-generated content is labelled — and traceable
When the platform generates or materially alters public content, it’s marked as AI-assisted and carries a provenance record, so nothing published in your name is a black box:
- A machine-readable marker on generated public assets, per EU AI Act Article 50(2).
- A stored provenance record for every published asset: model · prompt_template_version · source_assets · policy_version · human_approver.
- Human editorial review is logged — you can always show who approved what, and when.
Hard limits
What our AI will never do
The intended purpose is marketing, operations, and pharmacist-supervised product information. Full stop.
Out of scope by design
- No diagnosis, no clinical triage, no treatment recommendations.
- No health or medical claims invented on supplements or cosmetics.
- No advertising of prescription-only medicines to the public.
- No acting as a medical device or clinical decision-support tool.
A customer question that veers into clinical territory is handed to a human pharmacist — the AI is built to escalate, not to guess.
The claims firewall
Every outbound asset gets a policy decision
This is the part small brands and agencies can’t replicate. Before any post or ad can be scheduled or receive spend, it passes the claims firewall, which checks, in order:
- Product category → the country's current rule-set version
- Evidence / source for any claim being made
- Forbidden claims (disease claims on supplements, prescription promotion) and authorised claims only
- Required disclaimers and the approval tier for the risk level
It then records the evidence and the versioned decision. The AI can flag a problem or propose a compliant rewrite — but it can never declare something legally compliant. That stays with people.
How we describe it — honestly
Disclosure
Talking to an AI? You'll know.
Where your customers interact with an AI agent — on WhatsApp or in support — it identifies itself as AI and offers a clear handoff to a human pharmacist. No one is tricked into thinking a bot is a person. This meets the EU AI Act Article 50 disclosure duty for systems that interact directly with people.
Control
You stay the pharmacist in charge
- A stop / override control on every AI workflow.
- Trace logs you can inspect — what ran, on what input, with which approval.
- Scoped, revocable access: you grant named permissions and can pull them in one click.
- Change control on the models and rule packs we ship — improvements roll out reviewed, not silently.
Regulation
How this maps to the EU AI Act
The AI Act’s transparency obligations (Article 50) take effect 2 August 2026. We’re built for them, and we go further where health or real spend is involved:
| Requirement | In Farmacy OS |
|---|---|
| Disclose AI interactions (Art. 50) | AI agents self-identify + offer a human handoff |
| Mark AI-generated content (Art. 50) | Machine-readable marking + provenance record |
| Human oversight | Approval gates on every consequential action |
| Not a high-risk / medical-device use | Purpose limited to marketing, ops & supervised product info |
| High-risk-grade controls anyway | Documented purpose, limits, override, trace logs, change control |
Scope note
Accountability
Who's accountable for what
Clear lines keep everyone safe. Farmacy OS provides the controls, the labelling, and the firewall. The pharmacy remains the responsible advertiser and the controller of its customer relationship. And the law itself is validated by counsel, not by our AI.
Founder action · [F]
Name the pharma-advertising counsel who signs off each launch-market rule pack ([qualified pharma-advertising counsel]), and keep the Article 50 posture reviewed against the final EU guidance and Code of Practice as they’re adopted.