AI ethics in practice: why Europe needs more than the AI Act, according to a new Euractiv piece

[Grim Callum]

A recent Euractiv opinion piece argues that the EU AI Act, despite being the world's first comprehensive AI law, is not sufficient on its own to guarantee really ethical AI outcomes across Europe, since legal compliance and actual ethical practice are related but distinctly different things. The argument lands at a notable moment, with the Act's main obligations for high risk AI systems having become enforceable on August 2 this year, even as a separate Digital Omnibus package has already pushed some of the toughest Annex III requirements back to December 2027.

The core tension the piece identifies is a familiar one from other regulated industries. A checklist based compliance regime, however well designed, inevitably creates incentives to satisfy the letter of specific documented requirements rather than the broader underlying ethical intent behind them. Organizations can build technically compliant AI systems, complete with the required conformity assessments, risk management documentation, and human oversight provisions, that still produce actually harmful or unfair outcomes in practice if the actual organizational culture around AI development treats ethics as a paperwork exercise rather than a substantive ongoing commitment.

That gap matters more given how much discretion remains built into the Act's actual implementation. National competent authorities are still being designated in several member states even as enforcement powers have already technically begun, and the AI Office's own guidance documents continue evolving in response to real world edge cases regulators didn't fully anticipate when the law was originally drafted. A framework this new, still being actively interpreted and refined in real time, inevitably leaves meaningful room for organizations to satisfy formal requirements while missing the actual spirit those requirements were designed to serve in the first place.

The piece's broader argument is that genuine AI ethics requires embedding ethical consideration directly into an organization's actual development culture and decision making processes, not just its formal legal compliance function sitting somewhere in a legal or risk department. That means practical steps like diverse teams actually reviewing AI systems before deployment, meaningful mechanisms for affected communities to raise concerns and have them heard, and genuine internal accountability structures that operate independently of whatever the bare legal minimum happens to currently require. Regulation sets an important floor, according to this framing, but treating that floor as the actual ceiling of ethical ambition is exactly the trap Europe needs to avoid as AI Act enforcement continues maturing over the next several years

Ai Ethics In Practice Why Europe Needs More Than The Ai Actwww.euractiv.com

[HenryThierry]

The compliance versus genuine ethics distinction is a pattern that shows up in basically every heavily regulated industry eventually, financial services, pharmaceuticals, environmental regulation, all of it. Companies get very good at satisfying documented requirements specifically while sometimes missing the actual underlying purpose those requirements were originally designed to serve

[Gerrard]

Curious what specific enforcement mechanism the piece actually proposes beyond just general appeals to better organizational culture and internal accountability. Culture change without any real teeth or genuine consequences attached tends to be pretty fragile once it collides directly with actual commercial pressure and competitive urgency to ship products faster than careful ethical review would otherwise allow

[CMPunk_Fan]

The Digital Omnibus delay pushing some Annex III obligations back to December 2027 is a pretty significant detail that could easily undercut the whole urgency argument here. If the actual toughest requirements aren't even enforceable for another year plus, the ethics versus compliance gap this piece identifies may simply not have had enough real time to actually manifest yet in practice

[Wardlow]

What strikes me most is how early this critique is arriving relative to the Act's own enforcement timeline. Usually this kind of compliance versus genuine culture gap only becomes obvious years into a regulation's life once enough real world enforcement cases have actually accumulated to reveal the pattern clearly.

Raising it this early suggests either genuinely sharp foresight from people who've watched this exact dynamic play out repeatedly in other regulated industries before, or it could just be a fairly generic critique that gets applied reflexively to basically any new regulatory framework regardless of its specific actual merits and design. Hard to tell definitively which one this is without reading the full original argument in complete detail

[Ryan1]

Diverse teams reviewing AI systems before deployment sounds sensible in principle, but it's exactly the kind of recommendation that's properly easy to state as a nice principle and notably hard to actually implement meaningfully and consistently at real organizational scale. Curious whether the original piece gets into specifics about what that would actually concretely look like day to day rather than staying at the level of a general aspirational statement