Source: Interconnects (Nathan Lambert) — 2026-07-12
Summary
Nathan Lambert argues that within roughly six months, the US is likely to ban or indefinitely delay any open-weight model that meaningfully exceeds the capability level of GPT-5.5, Claude Opus 4.8, or GLM-5.2, citing reported White House discussions of a new executive order and what he characterizes as an Anthropic-led political campaign against Chinese open-weight labs.
Key Takeaways
- Central claim: regulatory action, not technical capability, is now the binding constraint on how far open-weight models can advance in the US market.
- Cites reported executive-order discussions that would require frontier labs to share models with government 30 days pre-release under a "protected frontier model" designation.
- Frames Anthropic's public advocacy against Chinese labs as a regulatory-capture dynamic, given Anthropic's own commercial stake in the outcome.
- Names DeepSeek and Z.ai's GLM-5.2 as the open models currently closest to closed-frontier parity — and therefore most exposed to a potential ban.