The European Union lawmakers have given a final approval to the
world's first major regulatory framework aimed at governing AI, marking a
pivotal moment for the tech industry. Garnering 523 votes in favor, against 46,
with 49 abstentions, this legislative achievement underscores the EU's
commitment to setting global standards in AI governance.
Under
the AI Act, machine learning systems will be divided into four main categories
according to the potential risk they pose to society. AI applications that pose
risks to citizens' rights, including biometric categorization systems based on
sensitive attributes, emotion recognition, social scoring and predictive policing
solely based on profiling will be classified as "unacceptable risk"
and will be strictly forbidden.
AI
systems that negatively affect safety or fundamental rights will be considered
“high risk”. Clear obligations are imposed on these high-risk AI systems,
identified for their potential harm across various domains like health, safety,
rights, environment, democracy, and law. These systems must undergo risk
assessments, ensure transparency, accuracy, human oversight, and provide
avenues for citizen complaints and explanations regarding their decisions.
General-purpose
AI (GPAI) systems, and the GPAI models will not be classified as high-risk but
will have to comply with transparency requirements and EU copyright law.
However, more potent GPAI models with potential systemic risks will face extra
requirements such as model evaluations, systemic risk assessments, risk
mitigation measures, and incident reporting. Moreover, artificial or
manipulated content like images, audio, or videos ("deepfakes") need
to be clearly labeled as AI generated so that users are aware when they come
across such content.
Regulatory
sandboxes and real-world testing will need to be established at the national
level, accessible to SMEs and start-ups. These measures aim to develop and
train innovative AI technologies before their market placement, ensuring
compliance with regulations and fostering responsible AI development.
The
regulation is expected to be formally adopted before the legislative session
ends. It will come into force twenty days after being published in the official
Journal, with full applicability after 24 months. Exceptions include bans on
prohibited practices (enforced in six months), codes of practice (in nine
months), general-purpose AI rules (in 12 months), and obligations for high-risk
systems (in 36 months).
Source: https://www.europarl.europa.eu/news/en/press-room/20240308IPR19015/artificial-intelligence-act-meps-adopt-landmark-law
