The adoption of the Artificial Intelligence Act by the European Parliament this week leaves the EU in the interesting position of both leading the field on regulating AI and playing catch-up at the same time.
The Act will undergo a last round of legal checks and be formally endorsed by the European Council of national leaders. This is expected to happen by the summer, with the law due to come into full effect by mid-2026.
As with EU data protection regulations, enforcement will fall in the first instance to authorities in the member states, but Brussels has reserved to itself responsibility for policing general purpose AI systems such as ChatGPT.
When the Act does come into force it will be the first serious attempt at a supranational level to harness the benefits of AI while putting limits on the potential downsides.
Parkinson’s: ‘I know it will beat me in the end, but I’m going to make it wait’
Mark Carney’s ‘back of envelope’ idea paved way for Ireland to renegotiate bank debt
Gracie Abrams in Dublin review: No nepo-baby here. A real star stands before us
All the president’s men: who might be in the Oval Office to meet the Taoiseach and what to watch
The EU has adopted a risk-based approach with some tools banned outright because of the dangers they pose. They are defined broadly as systems which threaten the rights of EU citizens and include products that manipulate human behaviour and exploit vulnerabilities. This catches mass surveillance and profiling systems, with some exemptions for law enforcement.
Below these come what are called high-risk systems which have a significant potential to do harm because of the tasks they carry out and the areas in which they are deployed, such as critical infrastructure and public service. They are subject to monitoring and human oversight.
Generative AI systems such as ChatGPT are subject to a separate set of rules regarding transparency and copyright for General Purpose AI systems. The more powerful versions of these products will be subject to further restrictions.
The advent of powerful generative AI tools is one of the reasons it has taken the EU five years to get to this stage. With more and more powerful versions of these systems on the way it is unlikely to be the last word. However, it is a good start and sets a global benchmark for the US, China and other states as they grapple with the problem.
