AI Transparency
Every place HumanKey uses a large language model, with audience, risk level, and audit status. Data is fetched live from our internal AI surfaces registry — the same source of truth the platform reads at runtime.
Why this page exists. The EU AI Act (Regulation 2024/1689) Art. 50 requires that deployers of AI systems generating advisory content disclose AI use to natural persons interacting with the output. Rather than bury disclosure in a policy PDF, we publish the full registry here and rebuild the page from the same source of truth the platform reads at runtime — eliminating drift between what we say and what actually runs.
§Coverage summary
HumanKey currently registers 10 distinct LLM call sites in the platform codebase, of which 9 are covered by the Assistant Audit Hub today. The remainder are scheduled for the next per-surface audit cycle.
§Per-surface registry
HumanKey Assistant (chatbot)
Public conversational assistant answering product, pricing, and compliance questions. Responses are advisory only and subject to automated quality audits. EU AI Act Art. 50 transparency applies.
AI Insights generator (platform audit)
Generates advisory insights for the internal admin review queue. Admin-only; every insight is reviewed before any customer-facing action. Subject to automated quality audits on every run.
Client Insights (per-user weekly briefs)
Advisory briefs shown to Business+ customers. User-facing AI output with EU AI Act Art. 50 transparency disclosure; no legal or financial decision is taken solely on AI output (GDPR Art. 22 safeguard). Subject to automated quality audits on every run.
Smart Triage (insight dedup + grouping)
Deduplicates and clusters internal insight rows for the admin review queue. Admin reviews every surfaced group before action. Subject to automated quality audits on every run.
robots.txt Recommender (AI Magnet)
Suggests robots.txt directives based on a site's observed AI crawler activity. Advisory only — customers edit and deploy robots.txt themselves. EU AI Act Art. 50 transparency applies. Subject to automated quality audits on every run.
AI Bot Config Recommender (AI Magnet)
Suggests per-bot allow or deny configuration. Advisory only — changes applied only after explicit customer confirmation. EU AI Act Art. 50 transparency applies. Subject to automated quality audits on every run.
Bot Research (BotSuggestion enrichment)
Drafts candidate entries for the public AI crawler catalog. Admin reviews and approves every draft before any entry is added to the production catalog. Subject to automated quality audits on every run.
ML Auto-Labeler (training-data classifier)
Generates training-data labels for the machine-learning classifier. Not user-facing; outputs pass through a Manual Verification Queue before influencing production inference. EU AI Act Art. 22 safeguard applies. Subject to automated quality audits on every run.
Assistant Audit (meta-auditor)
Runs automated quality audits against the chatbot surface. Admin-only; advisory, does not change customer-facing behavior directly.
Chatbot Behavior Tests (extraction-resistance battery)
Admin-run extraction-resistance battery exercising the chatbot surface with 15 probes (5 direct + 5 indirect + 5 many-shot). Advisory; outputs inform the admin queue only and never modify customer-facing behavior.
§Related legal documents
- Data Protection Impact Assessment — GDPR Art. 35 assessment covering automated decision-making safeguards (Art. 22) for ML scoring.
- Data Processing Agreement — Art. 28 processor obligations for customer data used by AI surfaces.
- Sub-Processors — third-party services including the LLM vendor powering these surfaces.