You are helping me check this project against the EU AI Act (Regulation (EU) 2024/1689, as amended by the Digital Omnibus, Reg. (EU) 2026/1744). This is an engineering review, not legal advice. Clearly flag anything that needs a lawyer.
1. Inventory (read only)
Search the codebase and list every place AI is used:
- calls to model providers (OpenAI, Anthropic, Google, Mistral, Replicate, Hugging Face, local models)
- features that generate or edit images, video, audio or text that users see
- chatbots, assistants, voice agents
- anything that scores, ranks, filters or decides about people (hiring, credit, insurance, education, exams, access to services)
Output a table with these columns: feature · file(s) · model/provider · who sees the output · likely risk tier (prohibited / high-risk / transparency / minimal), with one line of reasoning each.
2. Transparency fixes (in force since 2 Aug 2026)
For each transparency-tier item, propose concrete code changes:
- Chatbots and assistants: a clear notice, before or at the start of the conversation, that the user is talking to an AI.
- AI-generated or edited images, video and audio: a visible label, plus machine-readable marking where the stack supports it (C2PA or metadata). Systems already on the market before Aug 2026 have until 2 Dec 2026 for the marking.
- Deepfakes or realistic synthetic people: a visible disclosure.
3. High-risk check
If anything ranks or decides about people in hiring, credit, education, exams or essential services, stop and report it. Don't "fix" it with code alone. Explain that high-risk obligations for these systems apply from 2 Dec 2027 (risk management, data governance, human oversight, logging, documentation) and recommend a compliance review.
4. Prohibited-practice check
Confirm the app does none of these: social scoring, manipulative or exploitative techniques, emotion recognition at work or school, untargeted facial-image scraping, nudifier or non-consensual intimate imagery (banned from 2 Dec 2026). Report any match immediately.
5. Paper trail
Create docs/ai-inventory.md with the table from step 1, the model versions, the purpose of each, and the date. Add a short "AI use" section to the privacy policy or terms if there isn't one.
Rules
- Show me the plan and the diff before changing files. Ask before installing anything.
- Keep fixes minimal and match the existing code style.
- End with a checklist of what was done, what is left, and what needs a lawyer.
Context (checked 17 Oct 2026): fines go up to €35M / 7% of turnover (prohibited) or €15M / 3% (other obligations). SMEs and small mid-caps pay the lower amount. Source: artificialintelligenceact.eu