Negotiators in the EU are reportedly considering additional restrictions for large AI models, such as OpenAI’s ChatGPT-4, as a component of its coming AI Act.
Representatives in the European Union are reportedly negotiating a plan for additional regulations on the largest artificial intelligence (AI) systems, according to a report from Bloomberg.
The European Commission, European Parliament and the various EU member states are said to be in discussions regarding the potential effects of large language models (LLMs), including Meta’s Llama 2 and OpenAI’s ChatGPT-4, and possible additional restrictions to be imposed on them as a part of the upcoming AI Act.
Bloomberg reports that sources close to the matter said the goal is not to overburden new startups with too many regulations while keeping larger models in check.
According to the sources, the agreement reached by negotiators on the topic is still in the preliminary stages.
The AI Act and the new proposed regulations for LLMs would be a similar approach to the EU’s Digital Services Act (DSA).
EU lawmakers recently implemented the DSA, making it so platforms and websites have standards to protect user data and scan for illegal activities. However, the web’s largest platforms are subject to stricter controls.
Companies under this category, like Alphabet and Meta, had until Aug. 28 to update service practices to comply with the new EU standards.
The EU’s AI Act is posed to be one of the first set of mandatory rules for AI set in place by a Western government. China has already enacted its own set of AI regulations, which came into effect in August 2023.
Under the EU’s AI regulations, companies developing and deploying AI systems would need to perform risk assessments, label AI-generated content and are completely banned from using biometric surveillance, among other measures.
However, the legislation has not been enacted yet, and member states still have the ability to disagree with any of the proposals set forth by parliament.
In China, since the implementation of its AI laws, it has been reported that more than 70 new AI models have already been released.