The News: The AI Act is the flagship artificial intelligence regulation proposed for the EU. Formally proposed by the European Commission in April 2021, it aims to protect fundamental rights for the public, including nondiscrimination, freedom of expression, personal data use, privacy, and others.

The proposed AI Act divides AI applications into risk categories:

• Unacceptable risk – The law will prohibit uses for causing harm by distorting judgement based on exploiting vulnerabilities, biometric categorization based on sensitive or protected attributes, and general-purpose social scoring by public authorities.
• High risk – The law will allow applications categorized as high risk, but they will be heavily regulated and must meet requirements for risk management systems, responsible data governance, technical documentation, transparency, oversight, and accuracy.
• Minimal risk – Applications will have to meet transparency requirements.

The proposed AI Act has regulatory requirements for foundation models (large language models, diffusion, and other generative AI algorithms). Requirements include registry, performance levels, identify synthetic content, data governance, notify users of AI interactions, disclose if training on copyrighted data, and compliance documentation.

On June 14, the European Parliament voted to forward its position on the AI Act. This position serves as the European Parliament’s negotiating position with the two other governing bodies in EU legislation – the Council of the European Union and the European Commission. This EU legislative process is called the Trilogue negotiations. On July 23 a key negotiation, operational trilogue, will take place. Spain, which currently holds the EU Council of Ministers’ rotating presidency, will lead negotiations and wants an agreement in place by end of 2023. If this timeline holds, 2024 will be used to develop harmonized standards. By early 2025, the standards should be finalized and the AI Act will be in place as law.

You can read a summary of the AI Act’s standards process here.

Read a more in-depth summary of the current status here.

The EU’s AI Act: Q3 2023 Update

Analyst Take: While the foundations of the AI Act have been drafted by all three EU governing bodies, they were not all completely aligned. They will start with the European Parliament’s framework, but there is a lot that will likely change before negotiations wrap up, hopefully by the end of 2023.

• Friction, negotiation – The negotiations will include deciding on the definition of AI. As of now, there are three different versions. Another disputed area is classification of high-risk AI applications. Some entities believe most of the responsibilities under the Act will be with the vendors of general-purposeAI systems (Google TensorFlow, AWS SageMaker, etc.), with the “Know Your Customer” frameworks common in financial services and third-party audits used. Others, like the German AI Association, are advocating for very light regulation and more private sector initiatives.
• Sandbox idea – The European Parliament position includes a mandatory AI sandbox for testing and compliance for AI apps with regulatory supervision before general availability.
• Enterprise lobby – Large tech companies will engage directly with regulators, offering technical and development expertise. Google has engaged regulators, proposing watermarking to label AI-generated content to distinguish it from human-generated content.
• Advocacy groups – Because of AI’s potential far-flung reach, lobbying will come from the tech industry, but also industry associations, member governments, and civil groups.
• Worries about enforcement – The Brookings Institute believes implementation and enforcement have not been thought through sufficiently at this point and they see significant differences in the approaches from each of the three governing bodies. You can read the Brookings enforcement comments here.

Given the degree of difference in the three governing bodies proposals and the philosophic diversity of the lobbies, it would be difficult to say that the final AI Act will look a whole lot like the current proposed versions. The negotiation on July 23 will signal whether the biggest sticking points will be quickly resolved, or if it will require more time than 2023 to reach an agreement. One component that could delay an agreement is the regulatory requirements for foundation models. Governing bodies may hesitate to craft narrow legislation for this key generative AI element, given the nascency of generative AI. It is likely the EU’s timeline might be a bit aggressive, but The Futurum Group will reexamine how the negotiations are going and we will provide an update in Q4.

Disclosure: The Futurum Group is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.

Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of The Futurum Group as a whole.

Other insights from The Futurum Group:

UK AI Regulations Criticized: A cautionary tale for AI Safety

The Cost of The Next Big Thing – Artificial Intelligence

Managing the Challenges of Using the Contact Center Agent as the Human in the Loop

Author Information

Mark Beccue

Based in Tampa, Florida, Mark is a veteran market research analyst with 25 years of experience interpreting technology business and holds a Bachelor of Science from the University of Florida.