The European Commission has published a list of the first 100-plus signatories to the AI Pact, an initiative focused on getting companies to publish “voluntary commitments” about how they will approach and deploy artificial intelligence.
The EU's risk-based, legally binding AI rulebook (AI Act) came into force last month, but its compliance deadlines will take several years to come into full force, creating a non-compliance gap that the EU is seeking to fill with its AI treaty.
The initiative aims to increase engagement and foster commitment so that companies can take preemptive steps to implement the law's requirements ahead of the deadline. The agreement also focuses on facilitating information sharing, allowing signatories to cooperate with each other to address the upcoming requirements of the EU's AI rulebook and proactively develop best practices.
Additionally, there are three “core actions” that signatories to the agreement are expected to undertake (at a minimum).
Adopt an AI governance strategy to facilitate the adoption of AI within your organisation and comply with the AI Act in the future; Identify and map AI systems that may be categorised as high risk under the AI Act; Promote AI awareness and literacy among staff and ensure ethical and responsible AI development.
Beyond that, there is a long list of potential pledges (available here in PDF format) that was drafted by the AI Office, the oversight body for AI law, and filtered with feedback from “relevant stakeholders” in the AI Agreements Network. The resulting list of pledges essentially allows signatories to pick and choose which pledges work for them and combine them accordingly.
Examples include a pledge to “design AI systems that are intended to interact directly with individuals, and to provide appropriate notice to individuals that they are interacting with an AI system,” and to “clearly and identifiable label any AI-generated content, such as images, audio, or video, that constitutes a deepfake.”
This lengthy list is likely to encourage a compliance-focused competition among signatories to see who has the most to offer when it comes to AI safety.
Agreement to speed up compliance with AI law
The AI pact effort was announced by then-Internal Market Commissioner Thierry Breton in May 2023, with Google agreeing at the time to support the regulator's efforts. Now, more than a year later, the EU has many more signatories, but the list is missing some notable names.
Neither Apple nor Meta, for example, are on the list. The ad tech giant told Reuters on Tuesday it won't join immediately and wants to focus on complying with the AI law itself.
The penalties for violating the EU's legally binding AI rulebook are stiff, with fines of up to 7% of global annual revenues for violating prohibited uses of AI, up to 3% for violating other AI law obligations, and up to 1.5% for providing false information to regulators.
So if Meta gets it wrong with regards to the actual AI rules, they could be hit with billions of dollars in fines, which may explain why they've ignored the Pact so far, as breaking the pledge would only result in a public scolding.
Mistral, a major French language modeling company, is also not on the list – it's one of the AI law's fiercest critics, so it's not surprising that it has not signed the voluntary pledge either.
Meanwhile, Germany's Aleph Alpha, another major European language model maker, has also signed onto the agreement, but the company recently announced it would pivot to providing B2B support for its generative AI tools. Given its evolving business model, it too may be reconfiguring its policy priorities.
Other companies on the list include Amazon, Microsoft, OpenAI, Palantir, Samsung, SAP, Salesforce, Snap, Airbus, Porsche, Lenovo and Qulacomm.
Meanwhile, Anthropic, Nvidia and Spotify are nowhere to be found, the first two being particularly notable absences given the importance of AI development.
Spotify's absence is notable because the European company signed an open letter organized by Meta last week lobbying against regulations that could hinder the widespread adoption of generative AI.
A full list of EU AI Agreement Early Signatories can be found here.
Participating companies are a diverse mix, including large European telecommunications companies, consulting firms, software companies, banks/payments companies, multinationals, SMEs, consumer platforms, etc. Given how widespread and fast the adoption of generative AI technology is, it's clear that the 100+ names are just the tip of the iceberg.
Because these are entirely voluntary pledges, signing the AI Accord may amount to little more than a reputational boost: signatories will be asked to report on their progress 12 months after announcing their unique set of pledges, opening up another opportunity for publicity.