Artificial Intelligence & Machine Learning , Legislation & Litigation , Next-Generation Technologies & Secure Development

Europe's AI Act Poised to Become Law After Committees Vote

Two key European Parliament committees accepted a political compromise that aims to govern how trading bloc countries develop and deploy artificial intelligence.

See Also: GDPR & Generative AI: A Guide for Customers

Lawmakers from the civil liberties and internal market committees of the European Parliament approved the AI Act weeks after parliamentary negotiators and national state representatives had struck a deal (see: Europe Prepares for AI Act Enforcement).

"This regulation aims to protect fundamental rights, democracy, the rule of law and environmental sustainability from high-risk AI," the committees said in a statement. "At the same time, it aims to boost innovation and establishing Europe as a leader in the AI field."

The regulation will be the globe's first comprehensive AI regulation. Violations could cost companies up to 35 million euros or 7% of a corporate annual turnover.

The regulation bans high-risk high systems, such as emotion recognition, in the workplace or educational settings and social scoring, or scraping the internet for images used to train facial recognition algorithms. The regulation is still subject to approval by the full Parliament and the European Council - the body of direct representatives from national governments - but those steps are widely considered to be formalities.

The proposed AI transparency requirements, under which AI developers will have to be compliant with the existing cyber and copyrights measures, were also approved by an overwhelming majority. AI developers should give copyright holders the choice to opt out from adding their right-protected material from AI text and data mining systems, the regulation says.

Developers will be subject to transparency requirements and must comply with EU copyright law. The regulation also stipulates that AI systems should be adequately protected to avoid poisoning via datasets or training data.

The EU AI Act primarily seeks ensure cybersecurity through compliance with the newly passed Cyber Resilience Act, which stipulates that AI and other software vendors mandate software patching and updates. Under the Cyber Resilience Act, AI developers will have to disclose vulnerabilities to the European Union Agency for Cybersecurity.

To ensure privacy, AI systems must be compliant with the General Data Protection Regulation.

AI industry practitioners have decried copyright language, saying it could make it difficult to access books and other content required to train AI models.

"Good large language models need a lot of long documents in the training data. These rules are going to make training foundation models impossible," said Elliott Ash, assistant professor of data science and a machine learning researcher at ETH Zurich.

Since these datasets are likely to include copyrighted materials, current provisions in the EU AI Act could trigger litigation action, potentially preventing researchers from making "applications similar to ChatGPT," Ash said.

"Well-funded companies like Google and OpenAI are well insured by deep pockets to settle any copyright litigation, while MistralAI and academic researchers have to provide a documented data pipeline and may not have funds to settle copyright litigation."

The European Parliament is scheduled to vote on the regulation on April 11.