Sep 28, 2026 · 41 min · 9 segments
This episode offers essential insights on the EU AI Act and what it means for the future of trustworthy AI. Our guests unpack the benefits and challenges of risk-based regulation, ethical guidelines…
Sue-Ann TeoGuest
Therese MurphyHost
Fredrik HeintzGuest
It sounds like fascinating and very important work, tackling a range of challenges and seeking to build technologies that support European values, we could say, in some ways.




Then the challenge really becomes that from a practitioner's point of view, it leads to so much legal uncertainty.

And when the act was created, one of the arguments for it was that it was intended to reduce the legal uncertainty and provide a level playing field so that everyone knows the rules.

However, this is not the case in practice, and it will, as far as I understand, not be the case until it's actually been, I mean, ruled in courts.

And that will take, I don't know, five, 10 years, which means that we will now have a very long period of a large uncertainty where it's very difficult to get any clear answers.

So one of the things we discussed in terms of this language model training was you can take the Swedish king.

The Swedish king is a public figure, which means that then we are allowed to include information about the Swedish king.

And according to the GDPR and so on, if you're a private citizen, you have the right to not be.

Then we basically need to ask permission to include information about this person.

And now the question becomes, are you allowed to include information about the Swedish king or not? That's an open question.

It's more about, actually, I think one challenge is also the interplay between the AI Act and other regulation.

For example, GDPR has this data minimization requirement that you basically should collect as little data as possible.

While the AI Act says that you should have complete and accurate training data, which means that you should have as much information as possible.

It sounds like fascinating and very important work, tackling a range of challenges and seeking to build technologies that support European values, we could say, in some ways.




Then the challenge really becomes that from a practitioner's point of view, it leads to so much legal uncertainty.

And when the act was created, one of the arguments for it was that it was intended to reduce the legal uncertainty and provide a level playing field so that everyone knows the rules.

However, this is not the case in practice, and it will, as far as I understand, not be the case until it's actually been, I mean, ruled in courts.

And that will take, I don't know, five, 10 years, which means that we will now have a very long period of a large uncertainty where it's very difficult to get any clear answers.

So one of the things we discussed in terms of this language model training was you can take the Swedish king.

The Swedish king is a public figure, which means that then we are allowed to include information about the Swedish king.

And according to the GDPR and so on, if you're a private citizen, you have the right to not be.

Then we basically need to ask permission to include information about this person.

And now the question becomes, are you allowed to include information about the Swedish king or not? That's an open question.

It's more about, actually, I think one challenge is also the interplay between the AI Act and other regulation.

For example, GDPR has this data minimization requirement that you basically should collect as little data as possible.

While the AI Act says that you should have complete and accurate training data, which means that you should have as much information as possible.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.