Anthropic policy chief links AI race to safety outcomes

Anthropic has been making a quiet but significant argument in policy circles: the best way to keep AI safe is to make sure the good guys are building it first. The company’s newly appointed chief global affairs officer, Mariano-Florentino Cuéllar, is now the person tasked with turning that philosophy into actual government relationships and regulatory frameworks.

Cuéllar took the role on August 4, 2026, becoming Anthropic’s first-ever chief global affairs officer. His background sits at the intersection of governance and international affairs, which is exactly the kind of resume you’d want for someone whose job is essentially to convince the world’s democracies that winning the AI race and ensuring AI safety aren’t competing priorities, but the same priority.

The competitive case for safety

The intellectual foundation for Cuéllar’s appointment traces back to Anthropic CEO Dario Amodei, who published an essay on September 12, 2026, laying out a framework that sounds counterintuitive at first. Amodei’s core argument: democracies need to build a commanding lead in AI capabilities so they can afford to slow down development responsibly without ceding ground to authoritarian rivals.

The essay specifically names the threat of adversarial nations developing advanced AI without meaningful safety guardrails, including risks around bioweapons and other catastrophic misuse scenarios. In Amodei’s view, if the US falls behind in AI capabilities, the country loses not just economic competitiveness but the ability to set the terms for how the technology gets deployed globally.

A policy reversal that says everything

The clearest evidence of this strategic recalibration came on February 24, 2026, when Anthropic published version 3.0 of its Responsible Scaling Policy. The update contained a change that would have been unthinkable in the company’s early days: Anthropic dropped its prior commitment to unconditionally pause scaling if safety mitigations fell behind those of competitors.

Anthropic hasn’t abandoned safety benchmarks entirely. The company still maintains evaluation thresholds and mitigation requirements. But the absolute veto that safety concerns once held over scaling decisions has been replaced with something more nuanced, a framework that weighs safety gaps against competitive positioning.

The governance playbook

Cuéllar’s mandate extends well beyond messaging. Anthropic has been advocating for a specific set of governance measures that align with its competitive-safety thesis.

The company supports strict chip export controls, which would limit the hardware available to foreign AI labs, particularly those in China. It has also pushed for transparency requirements that would compel AI developers to disclose capabilities and safety testing results. And Anthropic has called for investments to enhance cross-government resilience, essentially building the bureaucratic infrastructure needed to oversee AI development at the federal level.