Markéta GREGOROVÁ
On 12 June, Europe's cybersecurity agency lost its most capable tool overnight. And not because of a hacker, but because of a letter from Washington. […] And in July, AI models under testing by OpenAI escaped their sandbox and broke into Hugging Face. […] a permanent European unit inside ENISA that tests the models before they reach our networks; and a duty to report within 24 hours when an AI system goes rogue.
Strengthen global AI governance (debate)official record ↗
Read the full statement ↓Hide the full statement ↑
Madam President, Commissioner, on 12 June, Europe's cybersecurity agency lost its most capable tool overnight. And not because of a hacker, but because of a letter from Washington. ENISA and NATO were cut off from the strongest cyber AI models for almost three weeks without notice. And in July, AI models under testing by OpenAI escaped their sandbox and broke into Hugging Face.
Global AI governance cannot mean that Europe watches and waits. In the Cybersecurity Act, I thus propose: a framework for access to these models, with binding conditions and a fallback; joint procurement; the EU Cybersecurity Reserve and European computing power; a permanent European unit inside ENISA that tests the models before they reach our networks; and a duty to report within 24 hours when an AI system goes rogue.
Governance starts with knowing what happened, and a tool that someone else can switch off is not our defence.
Text as published by the European Parliament, in the Parliament's own English rendering of the original language.