AI safety bottleneck is political will, not research — new analysis

In a new 29-minute read on the AI Alignment Forum, Charbel-Raphaël contends that the bottleneck for AI existential safety has shifted from research to political will. He defines a four-level pipeline for policymakers: awareness, conviction, engagement, and championship. His experience suggests the median senior policymaker starts below Level 0 (awareness that serious people see a big problem). He cites a striking data point: among 1,534 written submissions to the UN Global Dialogue, exactly one mentions “takeover,” and less than 1% reference existential risks.

He attributes this gap to structural imbalances: the US AI safety field employs roughly 3.6 researchers per advocate, many organizations self-censor, and funders treat repetition as redundant. Meanwhile, industry secured seven times more meetings with the European Commission on AI than civil society in 2023. The author urges prioritising advocacy, policy, and governance — echoing a February 2026 survey of AI safety leaders. He proposes shifting metrics from “clever papers” to “minds moved,” calling for immediate engagement before irreversible events occur.