FPJ Analysis: European Union's AI Act

The entry into force this month of the European Union (EU)’s Artificial Intelligence (AI) Act, to regulate general-purpose AI (GPAI) models, has been widely hailed a landmark. According to the law, the generality of AI models can be determined, among others, when models use at least a billion parameters and are trained with a large volume of data using self-supervision at scale and capable of performing a wide range of distinct tasks. A typical example of GPAI is a generative AI model that has in-built flexibility to generate content as text, image, and audio and video files. To demonstrate adherence to the onerous obligations, the law requires the high-tech industry to adopt a voluntary code of practice, an interim tool to promote compliance before harmonised European standards kick-in in 2027.

Safety, including health and fundamental rights concerns, transparency, copyright protection and independent monitoring and systemic risk mitigation of GPAI models form the cornerstone of the code. Specifically, signatory firms must furnish technical documentation to the AI office and competent national authorities, share information to players downstream to integrate their models into GPAI systems and evidence of in-house policies concerning copyright protection. GPAI models with systemic risk must report on the potential for serious incidents and accompanying corrective measures, besides ensuring cybersecurity protection.

Despite the US big tech’s general skepticism around the EU’s regulatory inroads, Google and Microsoft, among others, have expressed a readiness to sign Brussels’ AI code, while the social media giant, Meta, is a prominent holdout. A San Francisco federal court in June found that Meta’s use of millions of online books to train its AI models was consistent with the copyright law’s ‘fair use’ provisions. The judge nonetheless averred that the dilution of the market for copyright material consequent to Meta’s indiscriminate data mining might have served the plaintiffs (authors) as a stronger argument. The case, among numerous others, illustrates the skewed narrative in the AI debate wherein regulation is viewed as an impediment to innovation and competitiveness.

Mumbai Rains: BMC Pumping Stations Drain 16,451 Million Litres Of Water In 4 Days; Irla Leads With 3,768 ML (VIDEO)

In a shot in the arm for the champions of greater AI transparency and accountability, US senators overwhelmingly voted to remove a controversial amendment from President Trump’s One Big Beautiful Bill that would have imposed a 5-year freeze on state-level regulation of AI models and apps. A majority of American states are at various stages of enacting the AI legislation. Another bulwark against the EU’s endeavour to counter AI-induced distortion of basic human values is the legally-binding 2024 Framework Convention on AI and human rights, democracy and the rule of law. Adopted by the Council of Europe, the human rights watchdog that groups 47 European countries including all the EU member states, the Convention is avowedly technology-neutral, without undermining technological progress in any manner. The participation of non-member countries and multilateral bodies in the deliberations lends the Convention greater credibility.