How we use AI
Transparency about the models in this service, aligned to the EU AI Act's transparency expectations rather than written for marketing.
AI model transparency
To satisfy EU AI Act compliance, our service operates with full structural transparency concerning our underlying models. We document the specific capabilities, downstream limitations, and operational metrics of our AI deployment, providing a verifiable trail of model selection and processing logic to satisfy conformity assessments.
What the AI does
Our models analyse the structural requirements of the questionnaire, map the queries against your evidence, and generate initial drafts. To prevent errors, every draft response is strictly anchored to an exact text quotation found within your own documentation.
What the AI cannot do
The system is restricted from generating responses in the absence of source data. If a questionnaire item has no corresponding evidence in your files, the question is left unpopulated and flagged for your review.
Where inference runs
All model inference is executed within our secure EU (Ireland) environment via AWS Bedrock, keeping processing in-region. We enforce absolute data privacy, meaning your documents are completely excluded from any model training loops.
Human role
Users are required to thoroughly review the completed questionnaire prior to sharing it with any external buyer. This oversight obligation rests solely with you, a principle we outline with absolute transparency.
Prompt injection
We class all uploaded documentation as untrusted data. The files undergo strict normalisation and are pre-screened to detect and neutralise malicious instruction-shaped content before entering the model pipeline.
What we do not claim
We do not claim that AI-generated responses are infallible. To mitigate the risk of error, every draft is paired with an authentic source quote, ensuring that any missing evidence is highlighted for your attention rather than obscured.