Arthur Mensch, co-founder and CEO of Mistral, says all Artificial Intelligence vendors operating in Europe should pay a levy for the content they employ to train their models. He floated the proposal in an opinion piece in The Financial Times on March 20.
Paris-based Mistral has expanded rapidly over the past year and has positioned itself as the main alternative to U.S. Artificial Intelligence vendors, with a strong focus on data sovereignty for European states. This approach is paying off, with the company attracting significant funding, as seen in its $2 billion Series C funding round last September. Even so, Mensch argues that European vendors are not competing on equal terms with American and Chinese companies because of stricter and more fragmented copyright conditions in Europe.
Mensch said major U.S. and Chinese companies can train models domestically under looser or absent copyright rules, including on content sourced from Europe, while European developers face legal uncertainty. He also argued that the current opt-out framework, which allows rights holders to try to block the use of their content for training, is proving unworkable. In his view, that leaves Artificial Intelligence companies constrained by uncertainty while creators increasingly object to the unauthorized use of their work.
His proposed solution is a revenue-based levy applied to all commercial providers placing Artificial Intelligence models on the market or putting them into service in Europe, based on their use of content publicly available online. Mensch said the levy should also apply to providers based abroad so foreign companies contribute when operating in the region, creating what he sees as a level playing field within the European market.
Mensch said money raised by the levy would be directed into a central fund to support culture and new content creation in Europe, though he did not define the exact contribution level. Agence France-Presse subsequently quoted Mistral executive Audrey Herblin-Stoop, suggesting a rate of 1% to 1.5%. Mensch presented the idea as a starting point for debate and invited developers, policymakers, creators and rights holders to take part. Under current European Union rules, Artificial Intelligence companies can use copyrighted materials for text and data mining, including Artificial Intelligence training, unless a creator has reserved their rights.
