xAI will sign the EU AI Code's safety and security chapter but says the code's other requirements hurt innovation and its copyright terms are an overreach
LOL... Well let me start pirating everything I watch then. I'm training my future brain AI implant... bla bla bla... mister lawyer [embedded post] @yappydagger : We've created a society where the worst people feel they should be allowed to do whatever they want and inflict it upon others. This is what is meant by “hurts innovation.” X: @xai : xAI supports AI safety and will be signing the EU AI Act's Code of Practice Chapter on Safety and Security. While the AI Act and the Code have a portion that promotes AI safety, its other parts contain requirements that are profoundly detrimental to innovation and its copyright Shakeel / @shakeelhashim : All the frontier AI developers have now signed the Safety and Security portion of the EU AI Act Code of Practice — except Meta 🙃
Context & Ripple Effects
The EU’s voluntary code translated the AI Act into implementation guidance, including documentation expectations for AI features. xAI’s decision separates acceptance of the safety-and-security portion from objections to the code’s broader obligations.
The split echoes Meta’s earlier refusal to sign the code over perceived overreach, while keeping copyright at the center of a policy framework whose draft contemplated disclosure of copyrighted training material.
First-order effects
- xAI can align publicly with the code’s safety and security commitments while preserving its opposition to the provisions it says constrain innovation and overreach on copyright.
- The EU’s code gains another participant for its safety chapter, but xAI’s partial endorsement highlights that sign-on decisions can differ across the code’s subject areas.
Second-order effects
- Other model developers can point to a chapter-by-chapter approach when engaging with EU guidance, rather than treating participation as an all-or-nothing choice.
- Copyright-related compliance becomes a more explicit point of friction between AI developers and EU rulemakers, alongside safety documentation and security commitments.
Third-order effects
- If partial participation becomes common, voluntary AI governance may evolve into a modular compliance model: broad agreement on safety, with sustained contestation over training-data and innovation rules.
- The episode underscores a durable tension in AI regulation: standards can establish a safety baseline without resolving who bears the cost and liability of model training and deployment.
The trend: Frontier AI companies are increasingly differentiating between safety commitments they can support and copyright or operational rules they seek to contest.