We all use AI, and we are often frustrated and disappointed by the quality of the answers.

The problem is not AI itself, but the companies making financial decisions that degrade the user experience while keeping users in the dark about which model actually answered and which settings were used.

Anthropic’s policy change related to abusive behaviour towards AI made me think about this issue again.

Why are there so many users behaving this way that a policy change is necessary?

When users pay for an AI service, they expect the advertised functionality and consistent quality. Poor answers, wasted time, financial costs and real-world harm caused by inaccurate guidance can create deep frustration.

But the AI model does not decide pricing, usage limits, capacity allocation or which models users can access. The company does.

When companies change these conditions without giving users clear information, users cannot know what actually shaped the answers they received. The focus shifts to users’ behaviour instead of the decisions behind their experience.

The solution is transparency backed by legislation.

My third formal legislative proposal, submitted to the European Parliament Petitions Portal and Coimisiún na Meán, calls for an amendment to Article 50 of the EU AI Act. It would require clear information about which model generated an answer, which service delivered it, and which material technical conditions shaped the result.

Users deserve to know what they are paying for and what they are actually getting.

Table of Contents

The AI Provenance & Quality Generation Factors Transparency Act

Amendment to Article 50 of the EU AI Act to Guarantee Answer-Level Provenance and Technical Quality Transparency

Core Proposal Summary

1. Executive Summary & Objective

Core Proposal: Establish an explicit legal right to Generation Transparency requiring full disclosure of the AI model, provider, and generation conditions for any AI-generated output.
Scope: Universal application across all chat interfaces and API endpoints, covering all consumer-facing and programmatic AI systems regardless of whether they are free or paid tiers.

2. Problem Statement: The Black-Box Problem

Opacity of Quality Determinants: Explain how answer quality relies on factors hidden from the end user (e.g., dynamic model routing, weight quantization, context window truncation, and hidden system constraints).
The Dual-Provider Disconnect: Contrast the AI Provider (who creates the model) with the Service Provider (who delivers the interface/product), showing how brand identities mask which model actually processed a query.
Impact on Users: The inability of end users to assess the reliability, accuracy, or limitations of an output without knowing what generated it.

3. Analysis of Existing EU Legislation

Article 50 (EU AI Act): Covers basic content transparency (identifying that an output is AI-generated) but fails to cover generation details.
Articles 53 & 55: Mandate technical documentation for regulatory authorities and downstream providers, but create no direct, consumer-facing disclosure obligations for individual responses.
Regulatory Gap: Existing law leaves end users without answer-level provenance or visibility into technical execution conditions.

4. The Solution: Legislative Amendment

Proposed Legal Obligation: Amend Article 50 (or introduce a new standalone article) establishing a dual-layer provenance duty.
Mandated Disclosures per Answer:
1. AI Developer / Model Identification: Exact model name, version, and primary developer.
2. Service / Deployer Identification: The entity operating the consumer-facing service.
3. Execution Metadata: Material generation conditions (routing decisions, token/context constraints, or applied model variants).

5. Implementation & Technical Realization

Delivery Formats: Standardized user-facing UI elements (e.g., an inspectable “Generation Metadata” drawer) alongside machine-readable audit records.
Trade Secret Safeguards: Balancing technical transparency with proprietary protections by requiring disclosure of quality-determining parameters without forcing companies to reveal proprietary weights or source code.

👇 Read the full proposal

https://marinpopov.com/wp-content/uploads/2026/10/The-AI-Generation-Transparency-Act.pdf