A Big-Tent or Small-Tent AI Safety Movement?

Summary

Arvind Narayanan and Sayash Kapoor argue that the AI safety movement should become a “big tent” that includes people who reject or question AI existential-risk claims while still supporting serious work on catastrophic and systemic risks. They begin by agreeing with a broader diagnosis: societies chronically underinvest in preparedness, as shown by pandemic planning, cybersecurity’s limited attention to cascading failures, and gradual institutional erosion. In their view, AI can amplify these existing risks, but that does not make every amplified risk an alignment problem or justify treating superintelligence as an established existential threat. The authors say existential-risk warnings have nevertheless become the dominant public and policy frame because they are emotionally compelling, while more concrete AI-enabled cyber incidents and institutional harms attract less attention. They argue that this framing may be counterproductive by increasing polarization, encouraging panic-driven policy, and directing resources toward stopping future superintelligence rather than defending against harms from systems already in use. A generalized existential threat can also make a ban appear to be the only intervention, while bans would not address current catastrophic or systemic risks. The proposed big-tent movement would be pluralist in both beliefs and values, present positive public goals rather than panic, and pursue common policies such as transparency, embedded evaluations and liability for negative externalities. Stronger measures, including restrictions on fully autonomous recursive self-improvement, should be pursued by persuading skeptics rather than through policy-by-panic. The authors contrast this with a “small-tent” movement that prizes ideological homogeneity, deprioritizes risks viewed as merely catastrophic, and risks public backlash. They conclude by calling for a broader AI safety coalition.