Preparing for AI Act: Over 100 companies sign European AI pact

It is rare for companies to fulfill obligations in advance. Nevertheless, around 115 companies have already signed a European pact for artificial intelligence (AI). In doing so, they are voluntarily committing to complying with the principles of the EU's new AI regulation even before it takes effect. The participants include companies that develop large language models or solutions for generative AI themselves, such as Aleph Alpha, Google, Microsoft and OpenAI. The latter is surprising: last year, Sam Altman, head of the ChatGPT provider, warned loudly against overregulation by the European AI Act. He was annoyed that operators of high-risk systems would have to meet numerous requirements and carry out a risk assessment, for example.

Large AI users and service providers from the IT industry such as Adobe, Amazon, HPE, Lenovo, Samsung, OVHcloud, SAP, Cisco and IBM as well as the telco giants Deutsche Telekom, Orange, Telefónica and Vodafone have also joined the pact. There are also medium-sized companies from many sectors and corporations such as Airbus and Porsche. The controversial big data company Palantir from the USA is also taking part. The AI regulation prohibits certain areas of surveillance such as social scoring or the remote detection of emotions. Nevertheless, representatives of security authorities and business partners expressed confidence in the spring that the AI Act would not paralyze them in their core areas.

The pact was Breton's idea, future uncertain

The voluntary commitments include three core actions, such as following a regulatory strategy to encourage the adoption of AI in the organization and working toward future compliance with the AI Act. Systems that are likely to be classified as high-risk under the Act will be identified at an early stage. In addition, there is a commitment to expand employee expertise in order to "ensure ethical and responsible AI development". In addition, around half of the companies have promised to guarantee human supervision of the technology, mitigate risks and label certain types of AI-generated content such as deepfakes. By joining such self-regulation agreements, companies often also aim to avoid stricter requirements and, above all, sanctions.

The pact is regarded as the flagship project of Thierry Breton, the EU Internal Market Commissioner who has since resigned: the Frenchman once justified his initiative by saying that no time should be lost in the current AI race to ensure a safe online environment. According to him, Google was the first provider he discussed this idea with. It is unclear whether the new Commission will still attach great importance to the pact in the medium term, but it is still open to being joined. Names such as Meta, Anthropic and Mistral are still missing from the list. Meta said that it was still considering whether to join. The company welcomes harmonized rules and is focusing on implementation. The AI Act came into force on August 1. Some of its provisions are already applicable, but a transitional period of two years applies to many of them. Prohibitions take effect after six months, requirements for broad-based AI models after twelve.

(mki)