Large Language Models & Frontier Model Safety

Frontier models — systems above a defined training-compute or capability threshold — warrant a distinct regulatory tier because a small number of systems now have plausible large-scale misuse or failure modes (bio/cyber uplift, autonomous action at scale) that ordinary product-liability law was never built for. The party takes no position on whether AI itself is good or bad. The position is procedural: match the tier of oversight to the tier of capability, and keep the process public and predictable.

The June 2026 Claude dispute demonstrates the concern. Commerce Secretary Howard Lutnick ordered Anthropic to cut off Claude Fable 5 and Mythos 5 for every foreign national worldwide, including Anthropic's own overseas employees, over a discovered jailbreak vulnerability, on a model already running for hundreds of millions of users. Resolving it took eighteen to nineteen days of direct, undocumented negotiation between one company and one federal department: no public standard, no timeline, no appeal path. That's the pattern this issue exists to end.