Breaking News: Historic Deal Struck on the Artificial Intelligence Act, Paving the Way for Ethical AI.The European Parliament’s and the Council presidency’s negotiators have come to a provisional agreement. On the proposal for harmonised laws on artificial intelligence (AI), often known as the artificial intelligence act. After three days of “marathon” discussions. The purpose of the proposed rule is to guarantee the safety, respect for basic rights. And application of EU values of AI systems that are utilised in the EU and sold in Europe. Additionally, the goal of this historic initiative is to encourage AI innovation and investment in Europe.
Secretary of State for Digitalization and Artificial Intelligence in Spain, Carme Artigas
This is a significant historical accomplishment as well as a step forward! The deal reached today successfully tackles a major worldwide issue in a field. Where technology is developing quickly and is crucial to the survival of our economies and communities. And in the process. We were able to maintain a very tight balance that completely respects the basic rights of our citizens. While fostering innovation and the use of artificial intelligence throughout Europe.
Secretary of State for Digitalization and Artificial Intelligence in Spain, Carme Artigas
The AI Act is a premier legislative idea that might encourage both public and commercial players to develop. And use safe and reliable AI throughout the EU single market. The basic premise behind regulation of AI is to use a “risk-based” approach to control. Meaning that the more dangerous the technology is, the more stringent the regulations will be. Being the world’s first legal proposal of its sort. It has the potential to promote the European approach to tech regulation internationally by serving as a worldwide model. For AI regulation in other jurisdictions, much like the GDPR has done.
The principal components of the interim accord
The primary novel aspects of the preliminary agreement over the original Commission plan may be summed up as follows:
Regulations on high-risk AI systems and high-impact general-purpose AI models. That have the potential to create systemic danger in the future.
A revised governance structure that includes some enforcement powers at the EU level. An extension of the list of prohibited items. The ability for law enforcement to use remote biometric identification in public areas. All subject to safeguards and improved protection of rights. Such as the requirement that deployers of high-risk AI systems. Perform an impact assessment on fundamental rights before deploying an AI system.
More specifically, the following areas are covered by the temporary agreement:
Definitions and extent
The compromise agreement matches the definition of an AI system with the methodology suggested by the OECD. Guarantee that the definition offers sufficiently explicit criteria for differentiating AI from more basic software systems.
The interim agreement makes it clear that the rule does not apply to matters beyond the purview of EU law. With carrying out activities related to national security. Additionally systems that are solely utilised for military or defence objectives are exempt from the AI Act.
AI techniques that are forbidden and classified as high-risk
A horizontal layer of protection, including a high-risk categorization. Is provided by the compromise agreement to make sure that AI systems that are not likely to seriously violate basic rights or pose other major hazards are not detected. AI systems that pose very little danger would be subject to very minimal transparency requirements. Such as stating that the material was created using AI so that consumers may decide whether or not to utilise it further.
Numerous high-risk AI systems would be approved, but only after meeting certain conditions and fulfilling certain duties in order to be allowed to enter the EU market. The co-legislators have clarified and modified these requirements so that they are more technically feasible and less onerous for stakeholders to comply with. Examples of these adjustments include the requirements regarding data quality and the technical documentation that SMEs must provide to prove that their high-risk AI systems meet the requirements.
Changes to the compromise agreement define the duties and responsibilities of the different participants in the complex value chains that build and deploy AI systems, especially the suppliers and consumers of those systems. Additionally, it makes clear how obligations under the AI Act relate to those currently covered by other laws, such sector-specific or applicable EU data protection laws.
Certain AI applications are considered too risky, and as a result, the EU will prohibit certain systems. The provisional agreement prohibits a number of practices, including social scoring, the untargeted scraping of CCTV footage or internet photos of faces, emotion recognition in the workplace and in schools, biometric categorization to infer sensitive information like sexual orientation or religious beliefs, and some forms of predictive policing.
Excemptions for law enforcement
A number of modifications to the Commission’s recommendation regarding the use of AI systems for law enforcement were decided upon in light of the unique needs of law enforcement agencies and the necessity to maintain their capacity to employ AI in their essential job. These modifications are intended to reflect the requirement to maintain the confidentiality of sensitive operational data in connection with their operations, subject to the necessary protections. In an emergency, for instance, law enforcement authorities are now able to use a high-risk AI tool that failed the conformity evaluation process thanks to the introduction of an emergency procedure. But in order to guarantee that basic rights would be adequately safeguarded against any possible abuses of AI systems, a particular procedure has also been developed.
Furthermore, the provisional agreement makes clear the goals in cases where the use of real-time remote biometric identification systems in publicly accessible areas is absolutely required for law enforcement operations, and in those cases, law enforcement officials should be granted special permission to employ such systems. Additional protections are included in the compromise agreement, which also restricts these exceptions to situations involving victims of specific crimes, the prevention of real, imminent, or anticipated threats like terrorist attacks, and searches for suspects in the most serious crimes.
AI systems with a broad aim and foundation models
To address scenarios in which artificial intelligence (AI) systems are multipurpose and when such technology is later incorporated into another high-risk system, new requirements have been implemented. General-purpose AI (GPAI) system particular circumstances are also covered under the preliminary agreement.
Additionally,
Certain guidelines have been reached for foundation models, which are sizable systems that can proficiently carry out a variety of unique activities, including the creation of text, photos, video, and computer code as well as lateral language conversation and computation. According to the interim agreement, foundation models have to adhere to certain transparency requirements to be sold. An more stringent policy was implemented for “high impact” foundation models. These foundation models can spread systemic risks throughout the value chain since they were trained using a vast quantity of data and have advanced complexity, capabilities, and performance that is well beyond normal.
A fresh framework for governance
An AI Office inside the Commission is established with the responsibility of supervising these most cutting-edge AI models, promoting standards and testing procedures, and enforcing the common norms throughout all member states in response to the new regulations on GPAI models and the evident necessity for their enforcement at the EU level. The AI Office will receive guidance on GPAI models from a scientific panel of independent experts. This panel will help develop methods for assessing the performance of foundation models, provide guidance on the identification and emergence of high impact foundation models, and keep an eye out for any potential material safety hazards associated with foundation models.
Member states will have a significant role in the implementation of the legislation, including the creation of foundation model codes of practice, through the AI Board, which would be composed of members from the member states. The AI Board will continue to serve as a platform for coordination and advice to the Commission. Lastly, to offer technical knowledge to the AI Board, an advisory forum including members from business, SMEs, start-ups, civil society, and academia would be established.
Penalties
A fixed sum or a percentage of the guilty company’s global yearly revenue from the preceding financial year was used to establish the penalty for violations of the AI Act. This equates to €35 million, or 7%, for using prohibited AI applications, €15 million, or 3%, for breaking the AI Act’s requirements, and €7,5 million, or 1.5%, for providing false information. Nonetheless, in the event that the AI Act’s rules are broken, the preliminary agreement restricts administrative fines for SMEs and start-ups in a more reasonable manner.
The compromise agreement further states that a natural or legal person may file a complaint about non-compliance with the AI Act with the appropriate market surveillance authority and may anticipate that the complaint would be addressed in accordance with the authority’s established protocols.
Openness and safeguarding of essential rights
A fundamental rights impact assessment must be completed before a high-risk AI system’s deployers may place it on the market, according to the interim agreement. Furthermore, the interim agreement calls for greater openness with relation to the application of high-risk AI systems. Interestingly, a few clauses in the Commission plan have been changed to state that certain public bodies using high-risk AI systems would also need to register with the EU database for high-risk AI systems. Furthermore, recently added clauses emphasise that users of emotion recognition systems have a duty to notify natural persons when they are subjected to such a system.
Policies that foster innovation
The rules pertaining to measures in support of innovation have been significantly changed from the Commission proposal in order to provide a more innovation-friendly legislative environment and to encourage evidence-based regulatory learning.
Notably, it has been made clear that testing novel AI systems in real-world settings should be permitted in AI regulatory sandboxes, which are intended to create a controlled environment for their creation, testing, and validation. Additionally, new rules that permit testing of AI systems in real-world settings with certain restrictions and safety measures have been implemented. The preliminary agreement allows for some restricted and clearly defined derogations and contains a list of activities to be implemented in order to benefit smaller businesses by easing the administrative load on them.
Onset of force
With minor exclusions for certain clauses, the interim agreement states that the AI legislation shall take effect two years after it is enacted.
Next actions
The technical work to complete the provisions of the new legislation will continue in the upcoming weeks, following today’s preliminary agreement. After this process is finished, the presidency will offer the compromise text to the representatives of the member states (Coreper) for approval.
Until the co-legislators formally accept the document, it must first be validated by both institutions and go through legal and linguistic review.
Background data
A crucial component of the EU’s strategy to promote the creation and use of safe. legal AI that upholds basic rights throughout the single market is the Commission proposal, which was made public in April 2021.
In an effort to provide legal certainty, the proposal establishes a standard, horizontal legal framework for AI using a risk-based approach. The proposed rule intends to ease the creation of a unified market for AI applications, improve governance and the efficient implementation of current laws pertaining to safety and basic rights, and encourage investment and innovation in AI. It is closely related to other programmes, such as the coordinated strategy on artificial intelligence, which intends to increase European investment in AI. The Council decided on a general strategy (negotiating mandate) on this matter on December 6, 2022, and in mid-June 2023, interinstitutional negotiations, or “trilogues,” with the European Parliament began.
