Detection ruleset

The scanner runs on a maintained, versioned signature database — not a static prompt. This page is the public proof of that: every change to what we detect is logged here.

v1.0.0 — 2026-08-25 (initial)

• Chat platforms: Intercom, Crisp, Tidio, Drift, Zendesk, HubSpot Conversations, Freshchat
• Dedicated AI-bot platforms: Chatbase, SiteGPT, Voiceflow, Botpress, Landbot, Chatling, Botsonic
• Voice AI: Vapi, Retell, Bland.ai, ElevenLabs, PlayHT, Cartesia
• AI media generators: HeyGen, Synthesia, D-ID, Runway, Sora/Veo/Kling references
• Generic signals: LLM API references, chat-labelled UI elements, AI generator meta tags
• Obligation taxonomy aligned to Article 50 sub-clauses: interaction disclosure (chat + voice), synthetic content marking (image/audio/video), deepfake visible labeling, and the separate public-interest AI-text clause

Every category we check — check our work

Scams hide their detection logic behind a gauge. We publish our coverage. The scanner runs on a maintained, versioned signature database — the canonical file lives on GitHub. We do not publish the exact match patterns here, because that would let vendors trivially evade detection — but the full file is open source for anyone who wants to verify it.