US Adds Security Hurdle for AI Model Testing Abroad
The Biden administration has introduced a new requirement for sharing cutting-edge artificial intelligence models with international partners. According to a report by Politico, AI companies like OpenAI and Anthropic must now pass a U.S. government security review before granting Britain's AI Safety Institute (AISI) access to their latest models for evaluation.
The "America First" Review Policy
The directive, issued by the Office of the National Cyber Director, is rooted in a desire for the U.S. to conduct its own risk assessments first. The goal is to ensure these advanced AI systems pose no threat to domestic infrastructure before allowing allied nations to examine them.
This policy shift has immediate consequences. Anthropic has reportedly delayed providing its new Claude Mythos 5.1 model to the UK's AISI. The company stated it is coordinating with U.S. officials to gradually expand access to partners both at home and abroad.
UK's Stance and the Push for Global Standards
The UK's AISI has built established partnerships with leading AI firms, typically conducting safety tests before public releases. The new U.S. prerequisite disrupts this collaborative flow.
Britain's approach contrasts with the U.S. move. Prime Minister Andy Burnham has argued that AI safety is a global challenge requiring coordinated international action, not unilateral measures. The UK is actively campaigning to establish universal safety standards, positioning the AISI as a potential model for global governance.
Diverging Paths in AI Governance
The U.S. requirement emerges amid domestic debates on regulating rapidly evolving AI. Warnings from experts and companies about the potential dangers of insufficiently tested models have fueled these discussions.
Concurrently, the U.S. is bolstering its own oversight capacity. The Commerce Department's AI center is being strengthened to play a central role in domestic safety governance.
OpenAI advocates for a more collaborative path. The company suggested this week that U.S. agencies should work directly with international bodies like the UK's AISI to develop a cohesive global safety framework.
This adjustment in testing protocols is more than a procedural change. It highlights the growing tension between sovereign security priorities and the ideal of open international collaboration in the strategically vital field of artificial intelligence. The future of global AI governance will be shaped by such balancing acts.