The US government ordered Anthropic to suspend Claude Fable 5 and Mythos 5 for foreign nationals on June 12, citing national security export-control concerns over an alleged jailbreak. Anthropic disabled both models globally, leaving enterprises that had built automation on Fable 5 without their AI engine overnight. Within the same week, four open-weight coding models emerged as fallback candidates: Cohere's North Mini Code (Apache 2.0, single-H100 footprint), Moonshot's Kimi K2.7-Code (1T parameter MoE, long-horizon coding), Zhipu's GLM 5.2 (1M context, though benchmarks unverified), and Xiaomi's MiMo-V2.5-Pro. The key insight is that standardized agent tooling (Claude Code, Cline, OpenClaw) means swapping models requires only changing an endpoint, not rebuilding workflows. The episode demonstrates that enterprises with pre-qualified open-weight alternates kept shipping while others scrambled, turning model access into a supply-chain resilience question.
