Understanding the EU AI Act penalties and achieving regulatory compliance

The EU AI Act, expected to arrive mid-2024, prioritizes the ethical and responsible development and deployment of AI within the EU, seeking to safeguard fundamental rights and ensure public safety. Ushering in an era of responsible AI, the EU AI Act establishes stringent ethical and safety standards for AI systems, minimizing potential risks and maximizing societal benefit.

Financial penalties for non-compliance

To incentivize adherence, the EU AI Act prescribes significant financial penalties for non-compliance. These penalties can range from substantial fines to a percentage of global turnover, demonstrating the potential economic repercussions of failing to meet ethical and safety standards.

The AI Act outlines a tiered structure, with the severity of the fine directly linked to the nature of the violation. Member States have some flexibility in implementing the system, but the overall goal is to create a consistent and effective approach to penalties across the EU.

The cost of non-compliance

Even for large corporations, a hefty fine can significantly impact the bottom line. Non-compliance with the AI Act can lead to fines ranging from €7.5 million or 1.5% of turnover to €35 million or 7% of global turnover, depending on the infringement and size of the company. [1]

Estimating compliance costs

With the proposed AI act, an assessment of compliance costs is provided as a cost estimation of administrative burdens and detailed compliance costs. Developed by the Federal Statistical Office (FSO) of Germany (2a), this model accounts for the five regulatory requirements (2b) outlined in the European Commission White Paper on Artificial Intelligence. Specifically, it calculates costs based on key activities required for compliance with each standard.

Compliance Costs for an AI Unit

Regarding the compliance costs for an AI unit, these are centered around the aforementioned five requirements. Based on expert and industry analysis, and assuming developmental costs of €170,000 (2c), the cost breakdown is as follows: Training Data at €2,763; Documentation and Record Keeping at €4,390; Information Provision at €3,627; Human Oversight at €7,764; and Robustness and Accuracy at €10,733. Consequently, the estimated annual labor compliance cost for a single AI model is about €29,277. Extending these estimates to the global AI industry, the total compliance costs are estimated to fall between €1.6 and €3.3 billion, assuming that 10% of AI units are high risk and thus subject to these regulations (2d).

Certification costs

In addition to the annual labor compliance cost for a single AI product, the cost assessment also includes the expense of obtaining certification through conformity assessment of an AI model. Two estimates are presented: the cost of an ex-ante conformity assessment via an EU-type audit, and the cost of an ex-ante conformity report conducted by an internal Quality Management System (QMS).

The certification cost via EU-type auditing is estimated to range between €16,800 to €23,000, accounting for 10%-14% of development costs (2e). By contrast, establishing an internal QMS may cost between €193,000 to €330,000, with an annual maintenance cost of €71,400 (2f). It’s important to note that QMS costs can be distributed among multiple organizations to collaborate on building a QMS together; however, this approach may require significant alignment, planning, and business expenses.

Overall governance-related costs

The total governance-related costs for an AI model are about €52,227 per year, combining annual compliance costs of €29,277 and certification costs of €23,000 via an internal QMS or less expensive EU-type audit. This estimate is conservative, primarily covering development costs and not factoring in additional costs for frequent model retraining. A more comprehensive cost assessment would include an extra 1-4% additional cost for each retraining or change request.

Uncertainties in cost estimation

The above estimation is provisional, based on expert and market stakeholder insights. Several variables remain uncertain, including the final regulation details, the number of products classified as high-risk, and the extent to which businesses will rely on pre-trained AI systems versus developing their own.

Preparing for the EU AI Act

Despite these unknowns, businesses must prepare for these impending regulatory standards. Understanding the EU AI Act’s cost implications is imperative, but it’s only part of the equation: Organizations also need practical tools and guidance to effectively implement and maintain compliance.

2021.AI’s GRACE platform simplifies AI governance, risk, and compliance by reducing complexity and bureaucratic burden. GRACE’s pre-loaded assessments save time and effort, while its platform-agnostic integration seamlessly fits existing workflows. Its configurable options allow you to tailor your compliance strategy to your specific needs, minimizing financial and operational challenges.

Establish trust in AI

Build trust in your AI by implementing and monitoring AI governance, risk, and compliance across your organization. From model design to validation and certification, GRACE not only provides end-to-end alignment with the EU AI Act but also offers the following suite of options:

• Reducing complexity: Pre-loaded assessments, risk registers, and customizable workflows streamline the compliance process.
• Promoting seamless integration: GRACE works with any MLOps platform, fitting seamlessly into existing workflows.
• Enabling tailored compliance: Organizations can adapt their approach to their specific needs, minimizing time, effort, and cost.
• Building trust: Implementing GRACE fosters trust in your AI practices, boosting brand value and reputation.

While the impending arrival of the EU AI Act may present complexities, a thorough understanding of its penalty structure, coupled with strategic partnerships like the one offered by 2021.AI, can empower entities to navigate this evolving regulatory landscape. To gain further insights into how 2021.AI’s GRACE Governance can streamline compliance efforts for your organization, please contact us or book a meeting today.

Sources:

[1] EU Artificial Intelligence Act – political agreement reached on Friday 8 December https://www.ey.com/en_be/tax/tax-alerts/2023/eu-artificial-intelligence-act

[2a-2f] European Union (2021). Study to Support an Impact Assessment of Regulatory Requirements for Artificial Intelligence in Europe https://op.europa.eu/en/publication-detail/-/publication/55538b70-a638-11eb-9585-01aa75ed71a1 P. 11-12

You might also like…