A new set of AI rules is about to be implemented by the EU. Brussels has released a handy guidebook that will help companies understand what they can do and what they cannot.
On Wednesday, the European Commission announced the publication of the General-Purpose AI Code of Practice in order to help people comply with the AI Act. The Act allows parties to sign the code to show that they are in compliance. However, this is purely voluntary.
The Code is divided into three parts. It has two short chapters that outline the responsibilities AI companies have in terms of transparency as well as obeying copyrights. There is also a longer section on safety which the Commission deemed to be “relevant only to a limited number of providers of the most advanced models.”
The same goes for the AI Act as a whole. The AI Act, as we have noted in previous coverage, places the greatest burden of compliance on the most powerful and large frontier AI models. It also applies to those who operate in the most critical sectors, or whose operations are most likely to cause harm. The Code’s three sections include everything you’d expect in a typical AI regulation. For example, it prohibits the output of copyrighted material, requires AI scraping bots (AIs) to obey robots.txt (19459041), and mandates risk assessments.
The Code requires companies to complete a form detailing all the details of a model, including energy consumption dataand keep model documentation for a decade. It recommends that AI firms “also provide an online search engine” refrain from downranking any pages that refuse to ingest by their AI. And no, this is not directed at anyone in particular.
The Code is not yet official – it will need to be endorsed by the EU states and Commission. But even then, it won’t matter much. The EC has stated that there is no pressure on companies to sign up if they do not want to.
The Code does reflect the AI Act in its entirety, so companies ignore it at their own peril.
All of this assumes that the AI Act’s general purpose rules will be implemented on 2 August, as planned. A number of European firms have come out againstthe AI Act, calling for its simplification and delay to ensure continental AI companies would be able compete with unnamed “AI behemoths,” Some of whomhave also called for astop on enforcement.
CCIA’s European arm, a pro-tech trade group, expressed disappointment with the Code. They said it added confusion to a set rules that desperately need clarification. CCIA, for example,says that the Code imposes certain rules which are not required by the Act. Other requirements of the Act are also missing. Why would any company sign up for something so obvious?
- Meta’s latest plan to train AI using EU user data is still in violation of GDPR rules, according to noyb.
- EU rattles the purse and AI datacenters come running.
- ChatGPT’s launch polluted the planet forever, just like the first nuclear weapons tests.
- Google Cloud keeps AI data but not support in UK.
“Without meaningful improvements, signatories remain at a disadvantage compared to non-signatories, thereby undermining the Commission’s competitiveness and simplification agenda,” Boniface de Champris said. Others are urging EU to not bow to corporate pressure. The open letter was signed by a group of academics and privacy advocates who urged the Commission to not delay enforcement. This would undermine Europe’s commitment to its “putting consumer and fundamental rights at the center of all legislation.”
“We call upon the Commission to prioritise the full implementation and proper enforcement of the AI Act instead of re-opening or delaying its implementation,” principles. The Commission’s decision on AI Act enforcement will take some time, regardless of which way it swings. While the AI Act is effective as of next month the Commission will not have enforcement powers for some time. According to the EC “new models” models released after August 2, 2025 will have a year to become fully compliant. Older models will get two years. There could be a lot of changes between now and then. (r)
