Source: Financial Times (paywall)
Anthropic's decision to withhold Claude 5.1 from the UK's prerelease safety testing—a voluntary program designed to give regulators early access to frontier models—reflects a hardening stance on model distribution. It raises a concrete question: Do US AI labs see regulatory cooperation as shared governance, or as competitive liability? The UK frames this as US protectionism. The concern is real: if American labs share less while facing lighter domestic oversight, they gain speed-to-market relative to competitors in more heavily regulated jurisdictions. This choice shapes whether frontier AI development becomes a coordinated global practice or a fragmented race where regulatory engagement turns optional based on geopolitical calculation.