Poland's AI Systems Act Draft 2.0: Enhancements and Opportunities for Growth

On February 5, 2025, the Ministry of Digital Affairs published the second version of the AI systems act draft, partially incorporating feedback from public consultations, including input from AI Chamber. Unfortunately, the published documents do not include a table of submitted comments, specifying whether they were accepted or not and why. This lack of transparency makes it difficult to fully analyze the changes and assess their impact.

Nevertheless, a preliminary review suggests that the draft has been significantly improved, and many changes reflect, either directly or partially, the suggestions made by AI Chamber. There is a noticeable shift in emphasis and a softening of the initially highly restrictive approach.

The introduction of regulatory sandboxes for AI is a particularly positive development. This long-awaited step could create favorable conditions for the development of AI-based models and services in Poland. If implemented earlier than in other EU countries, it could give Polish businesses and the domestic market a competitive advantage.

However, while the new provisions are a step in the right direction, they will require further refinement to ensure that Polish entities have access to the most beneficial and accessible legal solutions in this area.

Remaining available to the Ministry for further legislative work, we present AI Chamber’s detailed comments on the AI Act draft from February 5, 2025:

Detailed Comments

1. Definition of AI systems
   We maintain our concern regarding the partial replication of definitions from Regulation 2024/1689. The chosen legislative technique raises serious concerns, and the justification for the draft does not explain why this approach is necessary.
2. Artificial Intelligence Development and Security Commission (KRiBSI)
   Despite some improvements, the structure of KRiBSI remains problematic. The legal status of the Commission’s Chairperson as a single-person central administration authority would better fit Polish administrative law traditions and prevent procedural inefficiencies. The current model, where KRiBSI is a collegial body, creates an unnecessarily complex legal framework.
3. Unclear selection criteria for KRiBSI members
   The draft does not explain the criteria for selecting institutions represented in KRiBSI. While the Commission’s composition has been reduced, which is a positive change, the selection process remains unclear. Excluding ministry representatives could be an attempt to increase KRiBSI’s independence, but in practice, it may disconnect the Commission from key decision-making bodies that influence AI development in Poland.
4. Risk of political misuse
   Article 55(2) outlines the criteria for issuing orders to withdraw or ban AI systems from the market. These criteria are highly subjective, yet such decisions can have severe financial consequences for businesses, potentially leading to bankruptcy.
   A particularly alarming addition allows for sanctions against AI systems that “influence or may influence public opinion and sentiment in a way that is significant for upcoming elections.” This criterion is dangerously vague and speculative, allowing for politically motivated bans. We strongly recommend removing this provision.
5. Legal inconsistencies in KRiBSI’s structure
   The legal construction of KRiBSI is overly complex. The Chairperson and the Commission are both considered administrative bodies, creating confusion over decision-making authority. The possibility of delegating powers from the Commission to the Chairperson suggests that the current structure is ineffective. A more centralized model should be considered for efficiency.
6. Flawed decision-making process
   The criteria for banning AI systems under Article 55(2) are highly discretionary. Granting immediate enforceability to such decisions poses a significant risk to businesses. Such regulatory powers should be exercised only in extreme cases and with clear evidence of harm.
7. Regulatory sandboxes and innovation support
   The introduction of Chapter 8: “Measures Supporting Innovation” is a highly positive step. The lack of innovation incentives was one of the biggest criticisms of the original draft. Establishing regulatory sandboxes was a key demand from the AI industry.
   However, for these sandboxes to be effective, they must be genuinely accessible. The admission process must be fast, low-cost, simple, and free from excessive discretion. The current provisions fail to ensure this, risking ineffectiveness. We recommend simplifying the admission process.
8. Conflicts of interest in KRiBSI membership
   The current conflict of interest regulations in Article 26(5) are too broad, allowing for potential political pressure on Commission members. The provision should clearly state that members must recuse themselves only in cases of direct conflict of interest.
9. Legal inconsistencies in citations
   In Article 12, the reference to Regulation 2024/1689 is reversed (listed incorrectly as 1689/2024). This editorial mistake should be corrected.
10. Unjustified expiration of Council member mandates
    We strongly oppose the provision in Article 25(6)(7), which states that Council members lose their mandate if the organization that nominated them withdraws their support. Council members do not formally represent their nominating organizations, so this mechanism for terminating mandates is unjustified. We recommend removing this provision.

Final Thoughts

The second version of the AI Act draft represents a significant improvement, incorporating many positive regulatory changes. However, several key issues remain unresolved, particularly in the structure of KRiBSI, the excessive discretionary powers granted to authorities, and the lack of clear selection criteria for Commission members.

While regulatory sandboxes and innovation support measures are welcome additions, their effectiveness depends on accessibility and procedural clarity. If the current bureaucratic barriers are not removed, these regulations risk being ineffective.

AI Chamber remains committed to further dialogue with the Ministry of Digital Affairs, advocating for a regulatory framework that fosters AI innovation while ensuring legal clarity and market stability.