AI Transparency Log

Last updated: 2026-10-01

What is this page?

This is a complete public list of every AI system Shower operates. Some are visible to you in the product; others run internally to build, review, and harden Shower itself. We list them all here so you can see exactly where AI is used, what data it processes, and whether the system is compliant with EU AI Act Article 50 transparency obligations (Regulation (EU) 2024/1689, hard activation 2 August 2026).

För dig på svenska: Detta är en komplett publik lista över varje AI-system Shower driver. Några ser du i produkten; andra kör internt för att bygga, granska och säkra Shower själv. Vi listar alla här så du kan se exakt var AI används, vilken data det behandlar och om systemet uppfyller EU:s AI Acts Artikel 50 transparenskrav (Förordning (EU) 2024/1689, hård aktivering 2 augusti 2026).

For more detail on how Article 50 maps to specific Shower features, see the full disclosure at /legal/ai-act-disclosures. A machine-readable feed of this exact same data is at /api/ai-transparency.

Customer-facing AI / Kundvänd AI

These AI systems produce output you can see or interact with. Each is annotated with a Shower badge in the product where possible (look for the "AI was used here" chip).
Dessa AI-system producerar utdata du kan se eller interagera med. Varje är märkt med en Shower-badge i produkten där det är möjligt (sök efter "AI användes här"-chippet).

customer-builder-chat

Customer-facing / Kundvänd

Compliant / Uppfyller krav

Purpose (English): Generates websites and applications from your prompts; the heart of Shower.
Syfte (Svenska): Genererar webbplatser och applikationer från dina prompts; kärnan i Shower.
Models / Modeller: anthropic/claude-opus-4-8
anthropic/claude-sonnet-4-6
Data processed / Data som behandlas: Customer prompt text
Customer-uploaded images
Customer project files (DSL, schema, copy)
Where used / Var det används: Customer-facing builder UI (/builder, /chat)
AI Act classification / AI Act-klassificering: Limited risk / Begränsad risk (Art. 50)
Article 50 marker / Markör enligt Artikel 50: Builder UI is overtly AI-driven (Article 50(1)): customers build by writing to an AI in a chat.

asset-generators

Customer-facing / Kundvänd

In progress / Pågående

Purpose (English): Generates and edits images, illustrations, icons, and other static assets used inside your project.
Syfte (Svenska): Genererar och redigerar bilder, illustrationer, ikoner och andra statiska tillgångar som används i ditt projekt.
Models / Modeller: google/gemini-3.1-flash-image-preview
photoroom/image-api-v2-edit (product photos)
Data processed / Data som behandlas: Customer prompt text
Customer-uploaded product photos and background photos (product photos only, sent to Photoroom)
Where used / Var det används: Asset wizard, generate-image endpoint and the builder's image tools (/api/assets, /api/generate-image)
AI Act classification / AI Act-klassificering: Limited risk / Begränsad risk (Art. 50)
Article 50 marker / Markör enligt Artikel 50: C2PA Content Credentials embedding pipeline targeted for July 2026 (Phase 3 per ai-act-disclosures.md §3).

composition-pre-step

Customer-facing / Kundvänd

Compliant / Uppfyller krav

Purpose (English): Classifies your prompt and routes it to the right specialised builder pipeline (e-commerce, portfolio, SaaS dashboard, etc.) before generation begins.
Syfte (Svenska): Klassificerar din prompt och dirigerar den till rätt specialiserad pipeline (e-handel, portfolio, SaaS-dashboard, etc.) innan generering börjar.
Models / Modeller: anthropic/claude-haiku-4-5-20251001
Data processed / Data som behandlas: Customer prompt text (only)
Auto-classified capability tags (generated, ephemeral)
Where used / Var det används: Pre-generation routing (src/lib/ai/chat-classifier.ts, complexity-inference.ts, persona-detector.ts)
AI Act classification / AI Act-klassificering: Limited risk / Begränsad risk (Art. 50)
Article 50 marker / Markör enligt Artikel 50: Routing decision is logged and surfaced to the customer in the builder UI; covered by general AI-driven-Service disclosure (Article 50(1)).

template-publish-gate

Customer-facing / Kundvänd

Compliant / Uppfyller krav

Purpose (English): Verifies your project does not leak personal data before it goes onto the public template marketplace; binary pass/fail.
Syfte (Svenska): Verifierar att ditt projekt inte läcker personuppgifter innan det publiceras på den publika mallmarknadsplatsen; binärt godkänt/ej godkänt.
Models / Modeller: anthropic/claude-sonnet-4-6
Data processed / Data som behandlas: Sanitised template manifest (DSL primitives + schema only)
Sanitiser whitelist diff
Where used / Var det används: Marketplace publish flow (post-whitelist sanitiser pass)
AI Act classification / AI Act-klassificering: Limited risk / Begränsad risk (Art. 50)
Article 50 marker / Markör enligt Artikel 50: Disclosed in ai-act-disclosures.md §8.2. Fail-closed posture: AI gate rejects on Anthropic outage; customer can revise + retry. No human review during the gate (lawful under GDPR Art. 22 because customer-data-only and customer-can-resolve).

published-app-assistant

Customer-facing / Kundvänd

Compliant / Uppfyller krav

Purpose (English): Answers the visitors of a customer's published app in the AI chat assistant the customer can add to it. What a visitor writes is sent to Anthropic, which writes the reply.
Syfte (Svenska): Svarar besökarna i en kunds publicerade app i den AI-chattassistent som kunden kan lägga till. Det besökaren skriver skickas till Anthropic, som skriver svaret.
Models / Modeller: anthropic/claude-haiku-4-5-20251001 (default)
Data processed / Data som behandlas: The last ten messages of the conversation: what the visitor wrote and the assistant's earlier replies
The assistant's name and instructions, written by the app's owner (the name cut to 50 characters, the instructions to 500)
Not saved in Shower's database; the chat window keeps the conversation only while the page is open
If the request to Anthropic fails, the error written to Shower's server logs can include these messages
Where used / Var det används: Published customer apps where the owner has turned on the AI assistant (chat window, /api/chat/published)
AI Act classification / AI Act-klassificering: Limited risk / Begränsad risk (Art. 50)
Article 50 marker / Markör enligt Artikel 50: The chat window shows "AI assistant" under the assistant's name and the line "You're chatting with an AI." directly under its header, in fixed colours that do not change with the colour the app's owner picks, from the moment it opens, before the visitor writes anything, and for the whole conversation; the message box points screen readers to that line.

Internal AI (engineering pipeline and admin tools) / Intern AI (teknisk pipeline och adminverktyg)

Most of these AI systems run inside Shower's own engineering pipeline (code review, planning, security audits, legal-doc sync) and do not handle customer data. Two do, and their cards say what: admin-ai-summarize sends customer prompt text to the AI for an internal admin dashboard, and ai-incident-responder reads error reports, which can contain customer details. We list them all for full transparency.
De flesta av dessa AI-system körs inuti Showers egen tekniska pipeline (kodgranskning, planering, säkerhetsaudit, juridisk dokumentsynkning) och hanterar inte kunddata. Två gör det, och deras kort säger vad: admin-ai-summarize skickar kunders prompttext till AI:n för en intern admin-dashboard, och ai-incident-responder läser felrapporter, som kan innehålla kunduppgifter. Vi listar alla för full transparens.