AI Regu­la­ti­on: Agree­ment bet­ween France, Ger­ma­ny and Ita­ly on foun­da­tio­nal models

-The AI Regu­la­ti­on is curr­ent­ly being Tri­lo­gue nego­tia­ted bet­ween the Euro­pean Com­mis­si­on, the Euro­pean Par­lia­ment and the Coun­cil of the EU. The Com­mis­si­on sub­mit­ted a pro­po­sal in April 2021, the Coun­cil adopted its posi­ti­on on Decem­ber 6, 2022 and the Par­lia­ment adopted its posi­ti­on on June 14, 2023. The first tri­lo­gue also took place on June 14, 2023, the second on July 18, 2023, the third on Octo­ber 2 – 3, 2023, the fourth on Octo­ber 17 and the fifth on Octo­ber 24.

The­re was par­ti­cu­lar dis­sent on

  • the Excep­ti­ons to the pro­hi­bi­ted prac­ti­ces for cri­mi­nal prosecution,
  • the Regu­la­ti­on of foun­da­tio­nal models (“basic models”) – Par­lia­ment wan­ted to pro­vi­de for regu­la­ti­on here;
  • Que­sti­ons of the Gover­nan­ce and enforce­mente.g. in the amount of the fines.

France, Ger­ma­ny and Ita­ly have now appar­ent­ly agreed – after a Report from Reu­ters – The “man­da­to­ry self-regu­la­ti­on through codes of con­duct” is sui­ta­ble for foun­da­ti­on models, in con­trast to “unte­sted stan­dards”. Foun­da­ti­on models are AI models that are trai­ned on lar­ge data sets and who­se results are gene­ral enough to be used for various tasks; this also inclu­des gene­ra­ti­ve systems such as ChatGPT. The AI regu­la­ti­on does not regu­la­te AI as a foun­da­tio­nal tech­no­lo­gy, but rather its use. Howe­ver, deve­lo­pers for foun­da­tio­nal models should have to pro­vi­de infor­ma­ti­on about their lear­ning model. This agree­ment should acce­le­ra­te the fur­ther cour­se of the debate.

An over­view of the glo­bal acti­vi­ties of AI regu­la­ti­on can be found at at the IAPP.




Rela­ted articles