AI Disclosure
Where Kernion uses AI, which models, what data they see, and what they cannot do.
Kernion uses AI as a core part of the Service. We treat that openly: this page is a plain-language statement of where AI is used, which models are used, what data they process, and what limitations apply. This disclosure is aligned with the transparency obligations in the EU AI Act (Regulation (EU) 2024/1689) for deployers of general-purpose AI systems.
1. Where AI is used on this site
- Chat widget on agency.kernion.io. An AI assistant that can answer questions about the product, features, and pricing. Your messages + a short system prompt are sent to a third-party inference provider under contract. No live connection to customer data.
- Text-yourself-a-demo. An opt-in public demo that sends an AI-generated SMS to a phone number you provide. Rate-limited. Suppression-list enforced. Phone number deleted after reply.
- Live-call theatre. A scripted replay of a real customer booking call. No live AI inference — it is a pre-recorded dialogue shown as evidence of a production deployment, clearly labelled "scripted demo".
2. Where AI is used in the platform (for customers)
- Voice + chat + SMS agent. Classifies intent, answers FAQs, collects booking details, escalates to a human when stuck.
- Content drafting. Drafts blog posts, FAQs, social posts, email replies, and site blocks. Every draft is a suggestion — the operator reviews before publishing or sending.
- AI search visibility (ASEO). Weekly probes across leading general-purpose assistants to track brand mentions and citations. No customer personal data is sent; probes use your public brand name and website.
- Compliance drafts. Drafts of privacy, imprint, and cookie notices from your declared facts + jurisdiction. Flagged as provisional until a human reviews and publishes.
3. How AI is sourced
We operate as a deployer of large language models under the EU AI Act. We do not train foundation models ourselves; we contract with third-party inference providers, select them on security, performance, and data-handling posture, and renegotiate the contract periodically. Providers are contractually restricted from training on API traffic.
We may change providers to improve quality, price, or latency. Named providers, region, and DPA references are provided to contracted customers under NDA on request to hello@kernion.io. Material changes are announced in our changelog.
4. What the AI cannot do
- It does not send live messages without operator approval. An outbound customer message goes through a platform-owned queue with suppressions, rate limits, and visibility to the operator.
- It does not execute financial, booking, or data-changing actions autonomously in the Core plan. Paid plans can opt into "act-with-confirm" or "auto" tiers per feature; those settings are explicit and reversible.
- It does not train on your data. Inference providers are contracted to not train on API traffic; this is documented in their terms.
- It is not a lawyer, doctor, accountant, or tax adviser. AI-drafted legal, medical, or financial output is a starting point, not advice.
5. Human oversight
Every AI surface has a reachable human: the operator (you, for your tenants), our support inbox (hello@kernion.io), and your supervisory authority for data-protection concerns. AI output can be wrong; operators are expected to review before acting.
6. Your rights regarding AI processing
- Access & transparency. Ask what data is being processed, by which model, in which feature.
- Objection. You can object to AI-assisted processing of your data under Art. 21 GDPR.
- Opt-out of AI training. We do not send your data for training. If you want written confirmation, request it through Data Requests.
- Human review. If an AI-assisted decision materially affects you, you can ask for human review.
7. Security of AI traffic
AI requests travel TLS-encrypted to the model provider. We use API-scoped keys rotated on a schedule, and we log request metadata (model, feature, latency) but not the prompts or outputs themselves unless operator-facing logging is opted in for debugging.
8. Contact
AI-related questions: hello@kernion.io.