The AI Safety Movement: Big Tent or Small Tent?

Introduction: An Overlooked Underlying Divide

Public debates in the AI safety field tend to revolve around specific issues—model alignment, regulatory pathways, open-source risks, the prioritization of existential threats. But beneath these surface-level arguments lies a divide that is rarely made explicit: should the AI safety movement be a big tent, or a small tent?

This divide is not merely a question of organizational strategy. It determines who gets to be called an "AI safety researcher," which issues deserve resources, and whom the movement ultimately seeks to influence.

Big Tent Logic: Inclusion in Exchange for Influence

The core argument of the big tent strategy is this: the challenges facing AI safety are severe enough that no one who might offer solutions should be excluded. This means welcoming participants from diverse backgrounds—reinforcement learning researchers, policymakers, ethicists, even technologists who are optimistic about AI capability development.

Supporters' reasoning is straightforward: if the AI safety movement is too narrow, it will lose its connection to the mainstream AI research community and, consequently, its real-world influence. A "pure" movement endorsed by only a few dozen people cannot play a role in key decisions. A big tent means more funding, more talent, and broader public discussion.

This approach is especially common at the policy advocacy level. When facing legislators or corporate decision-makers, a coalition encompassing diverse voices is clearly more likely to secure a hearing than a small, internally homogeneous group.

Small Tent Logic: Focus in Exchange for Depth

The small tent strategy emphasizes that when the definition of "safety" is expanded indefinitely, it risks becoming hollow. If any concern about AI can be folded into the safety agenda, then that agenda loses its analytical rigor and action priorities.

Small tent supporters worry that excessive inclusiveness leads to issue dilution—existential risk, algorithmic bias, labor displacement, and privacy violations get lumped together, and none of them receives adequate attention. More seriously, a big tent may bring in participants who fundamentally conflict with safety goals: those who are actually accelerating capability development while only talking about safety as an afterthought.

Another argument for the small tent is signaling clarity. When a movement has clear boundaries and core positions, it is easier for outsiders to understand where it stands. Blurry boundaries, by contrast, allow the real threats to be obscured.

Issue dilution is a classic concept in social movement research. It refers to the phenomenon whereby an organization or coalition's agenda continuously expands to incorporate more and more loosely connected demands, causing its original core positions to lose mobilizing power and policy traction due to the dispersion of attention and resources. Political scientists typically associate it with "issue creep" and "coalition overload": the larger the coalition, the smaller the intersection of internal preferences, and the resulting position papers tend to remain at the most abstract and vague level. In the AI safety context, the specific risk of issue dilution is that existential risk, algorithmic bias, privacy, and labor displacement each have different time scales, causal mechanisms, and audiences. Packaging them all under a single "safety" label means none of them can achieve sufficient expert density and funding intensity.

Signaling clarity draws on signaling theory from economics and game theory: the vaguer the positions a group broadcasts externally, the harder it is for outside observers (funders, legislators, media, potential new members) to judge its true preferences and strength of commitment, and the higher the screening cost of cooperation. The small tent emphasizes boundaries because clear identity definition is itself a commitment mechanism—it tells the outside world "what we truly care about," thereby attracting those who genuinely identify with that core goal rather than mere free riders. The cost is that the clearer the signal, the narrower the audience that can receive and respond to it.

The Essence of the Divide: Who Defines "Safety"?

Behind this tent-size debate is a contest over the right to define "safety." The big tent tends to keep the definition open so that new issues and new members can be absorbed at any time; the small tent tends to first establish core problems and then build a community around them.

Both strategies come with costs. The big tent faces coordination costs: fundamental disagreements among members can lead to internal friction, and may even prevent the movement from forming a unified position on key issues. The small tent faces marginalization risk: an overly narrow agenda may be seen as extreme or out of touch, losing its channel for dialogue with the mainstream.

One detail worth noting: this divide is not either-or. In practice, many organizations oscillate between big tent and small tent—expanding coalitions for external advocacy, maintaining focus for internal research. But oscillation itself has costs: outsiders may find it hard to judge the movement's true position.

Practical Implications for Participants

For researchers, engineers, or policy professionals wanting to enter the AI safety field, this tent debate directly affects their path choices.

Joining a big tent organization may mean richer interdisciplinary exchange, but one may also find one's specific concerns drowned out by grander narratives. Joining a small tent group may provide more concentrated discussion and a clearer sense of identity, but one may also find limited resources and difficulty scaling influence.

For funders, this question is equally critical: should funding go to general-purpose projects broadly covering multiple safety issues, or should it concentrate on a few core problems deemed most urgent?

Conclusion: The Divide Itself Needs to Be Acknowledged

There will be no permanent answer to the AI safety movement's tent-size debate. As technology develops and public discussion evolves, the movement's boundaries will keep shifting. But making this divide explicit is far healthier than letting it shape every resource allocation and recruitment decision through implicit assumptions.

Both big tents and small tents have their rationale and limitations. The real question is not which is better, but which strategy is more appropriate at what stage, for what audience, and in pursuit of what goals.

Background Note

What is really at stake here is a question of boundary work. Sociologist of science Thomas Gieryn used the term to describe how disciplines or professional groups claim authority, allocate resources, and distinguish "insiders" from "outsiders" through boundary-drawing. The AI safety movement is likewise engaged in boundary work: the big tent expands boundaries outward to secure legitimacy and resources, while the small tent tightens boundaries inward to preserve professionalism and judgment. Who has the right to define "what counts as an AI safety issue" is essentially who has the right to allocate attention, funding, and discursive power in this field. Understanding this reveals that the tent debate is not a matter of stylistic preference, but a structural conflict over where authority lies.

One-line Summary

The big tent vs. small tent debate within the AI safety movement is fundamentally a contest over the right to define "safety."

Paragraph Summary

The article points out that beneath public debates in AI safety lies a rarely articulated underlying divide: should the movement be an inclusive big tent or a focused small tent? The big tent trades inclusion for influence, funding, and talent, but faces issue dilution and coordination costs; the small tent trades focus for analytical depth and signaling clarity, yet risks marginalization. The essence of this divide is a contest over the right to define "safety," determining who can be called an AI safety researcher, which issues receive priority, as well as researchers' path choices and funders' capital allocation. The author argues that both strategies have their appropriate contexts, and the key is not which is better, but making this divide explicit rather than letting it steer decisions through implicit assumptions.