Gilad Abiri

The compliance pathway isn't clear, and that's precisely what makes this case so significant for the future of AI development in Europe.

The catch? All three cannot be maximized simultaneously.

This case demonstrates why GDPR, not the AI Act, poses the more fundamental challenge to LLM development in Europe. While the AI Act focuses on deployment and use, GDPR questions the legality of current training methods themselves.

OpenAI's response is telling. By arguing that the fine "undermines Italy's AI ambitions," they've highlighted the regulatory trilemma facing European authorities:
- maintain strict data protection standards
- preserve technological sovereignty
- remain competitive in AI development

This isn't unique to OpenAI - it's inherent to how contemporary LLMs are built, including European models like Mistral's. The core issue isn't technological but legal: the fundamental tension between GDPR's conception of personal data protection and current LLM development methods.

The case exposes a structural problem at the heart of large language model development. The Italian authority found what we've long suspected: OpenAI processed personal data for ChatGPT's training without an adequate legal basis under GDPR.

🧵 Have just started teaching GDPR in my Data Privacy class amid significant news: Italy's data protection authority has levied a €15 million fine against OpenAI for violations of GDPR in ChatGPT's operations.

ISP Publishes Collection on Artificial Intelligence and the Digital Public Sphere A new collection of essays explores how the advent of artificial intelligence stands to impact the digital public sphere.

"Artificial Intelligence and the Digital Public Sphere" is out now! Edited by Elisabeth Paar and @giladabiri.bsky.social, the essays in this collection show how AI systems are reshaping the fabric of our shared digital space.

law.yale.edu/yls-today/ne...

In conclusion, by inviting civil society back to the table, we aim at a balance: preserving the essence of public discourse without excessive state interference. The platform federation era needs new rules and limits. Old lessons from federalism might provide the answer.

By doing so, we allow civil society organizations to define public discourse norms. Just like old times, they can filter out hate speech, misinformation, and more. This ensures a civil digital realm, minus the heavy hand of the state. /8

Our proposition? Embrace "civil society federalism." Instead of the state singlehandedly policing the digital public sphere, platforms should reintegrate civil society into their gatekeeping fold. /7

While state federalism might offer a solution to digital chaos, it also threatens to impose state-driven values onto citizens. This risks undermining the liberal essence of democracies. Remember: there was a reason why we wanted states out of speech in the first place. /6

Democracies are replacing traditional gatekeepers with a nuanced approach. We call it state federalism. Some, like Germany, are pushing their civility norms by mandating platforms to ban any speech forbidden by criminal norms, while others, like France, are outlawing fake news. /5

Nations everywhere are grappling with this "bypass effect," striving to uphold their local speech norms. Autocracies might opt to entirely fence off their digital realms, but such drastic measures aren't feasible or even desirable for democracies. /4

With social media, there's a "bypass effect." Now, individuals can sidestep these gatekeepers, speaking directly to vast audiences. All good, but who is going to make sure we can meaningfully talk among ourselves now? /3

Until recently, media elites acted as the gatekeepers of mass communication, enforcing the social norms that enabled civil public communication. If anything, this system had the advantage of allowing mass communication while keeping the state out of unduly controlling expression. /2

Public conversation relies on shared civility norms. No public debate worth its name can happen among insults, misinformation, and hate speech. This is not a normative aspiration, it’s a fact: it just can’t happen. /1

Global platforms dominate our public discussions, and yet our political debates remain stubbornly local. Speech crosses borders, but the impact of speech remains local.

Our (@sebastianguidi) forthcoming paper in @YJoLT explores this “Platform Federation” 🧵​

papers.ssrn.com/sol3/papers....

Until recently, media elites acted as the gatekeepers of mass communication, enforcing the social norms that enabled civil public communication. If anything, this system had the advantage of allowing mass communication while keeping the state out of unduly controlling expression

Public conversation relies on shared civility norms. No public debate worth its name can happen among insults, misinformation, and hate speech. This is not a normative aspiration, it’s a fact: it just can’t happen.

I'm developing a seminar syllabus on generative AI's impact on online governance. Seeking essential readings, case studies, and resources. Recommendations are deeply appreciated.