Français
There's no such thing as "mostly open."
That's the point Thomas Vaughan, principal engineer at the Common Crawl Foundation, kept coming back to at POSAIS 2026. He joined our roundtable “Open and lawful: building generative AI Europe can trust”, and he didn't leave much room for nuance on what "open" actually means.
« If you want something to be trustworthy, safe and open, every single component of it has to be open. » Not the weights. Not the weights and the data. EVERY COMPONENT. Weights, training data, training code, the whole chain, or it doesn't count.
Common Crawl Foundation has spent close to two decades archiving the open web, freely, for anyone to use. It's the kind of infrastructure most people don’t think about, and the kind that most open AI models, including our own, depend on
Which is exactly why Tom's closing point landed the way it did: "𝗪𝗲 𝗺𝘂𝘀𝘁 𝗱𝗲𝗳𝗲𝗻𝗱 𝘁𝗵𝗲 𝗰𝗼𝗺𝗺𝗼𝗻𝘀. Common knowledge is knowledge that belongs to everyone, and it is under threat."
Not a slogan. A warning from someone who has spent his career maintaining the thing being threatened...
