Article 50-correct disclosure patterns
Every conversation carries the correct Article 50(1) and 50(2) disclosure for the matching jurisdiction: at the right moment, in the right language, with the sequence recorded.
Research programme
Compliance signal that the open web does not contain.
In development · first licence target Q2 2028
Synthetic, compliance-labelled voice and chat conversations, generated under Audact policy bundles and annotated for disclosure, scope, refusal and outcome.
- Opt-in only
- Template-generated, not extracted
Contents
Five properties a buyer of compliance signal actually needs.
Compliance-labelled outcomes
Each turn is tagged: pass, redact, refuse, escalate, scope violation or out of policy. The labels come from the live policy bundle, not from hand annotation.
Multi-jurisdictional coverage
NL, DE, FR, IT and ES at v1, expanding to all 27 member states. The same scenario is rendered against per-country overlays, so a model can learn the difference between jurisdictions rather than one local norm.
Multi-vertical scenarios
Dental, legal, financial services and healthcare communications — the verticals where compliance behaviour diverges most sharply from a generic open-web baseline.
Volume on demand
Synthesis is template-driven rather than extracted from a real customer. Volume can be generated without touching live personal data or live conversations.
Privacy
Synthetic does not mean laundered.
The dataset is generated from templates and policy bundles. That is a different thing from anonymising real conversations, and it is the reason the programme can exist at all.
- Opt-in through the data processing agreement. No customer conversation enters the dataset without an explicit amendment. Status: Planned.
- k-anonymity of at least 50: every record class contains at least fifty indistinguishable members before it can be licensed. Status: Design rule.
- No personal data: names, identifiers, voice samples and raw transcripts are excluded by design. Status: Design rule.
- External counsel reviews every licensing deal before signature, and the review log is retained.
Licensing
Three commercial paths, from sampling to a bulk licence.
Per data-pair
Confirmed at launch
- Sampling tier
- Priced by annotation depth and jurisdiction coverage
- For evaluation before a commitment
Bulk licence
On request
- Per provider, per year
- Includes regeneration as the policy bundle evolves
- New jurisdictions included as they are added
Annual subscription
Custom
- Continuous data stream
- Sized to the provider
- Commercial discussion through research@audact.ai
Commercial terms are confirmed at the standalone launch. Until then, access runs through the research programme.
Timeline
Where the programme stands.
2027
Planned
Building the dataset. The generation pipeline goes into production and the jurisdictional overlays expand from five to twenty-seven.
Q2 2028
Target
Target for the first licence deal: one EU model provider, on the bulk-licence tier.
Q4 2028
Target
Standalone product launch on public commercial terms.
Why it matters
There is no public-web analogue.
A compliance-labelled, EU-jurisdictional corpus of governed-conversation signal does not exist on the open web, because the labels come from a policy engine that was running at the time. That makes it useful for training, evaluating and benchmarking systems against real EU compliance requirements — and it is the reason this is a research programme rather than a product listing.
Next step
Early access for model providers.
The programme is in development. If you want to be on the list for the first release, write to research@audact.ai and tell us which jurisdictions and verticals you need.
Related
Data governance The aggregate research dataset and how access is granted.
The platform The policy engine the labels come from.