Is the quantum computing race a worse moral hazard than the AI race

[Warren]

Both the AI race and the emerging quantum computing race share a familiar, uncomfortable underlying structure, multiple well resourced actors racing hard to be first to a capability with enormous potential upside, under real competitive pressure that actively discourages slowing down to carefully consider safety, oversight, or broader societal consequences along the way. Whether one of these two races carries genuinely worse moral hazard than the other is worth thinking through carefully rather than simply treating them as interchangeable instances of the exact same generic problem.

The case for AI being the more concerning race rests on its sheer breadth and immediacy. Advanced AI systems are already deployed today across an enormous range of domains simultaneously, and the potential harms span an unusually wide spectrum, misinformation, labor displacement, autonomous weapons, and in the most extreme framings, genuinely existential risk to humanity as a whole. The race dynamic here is already fully live and actively unfolding in real time as we speak, right now.

The case for quantum computing carrying comparably serious or possibly even worse moral hazard rests instead on the specific binary, all or nothing nature of one particular payoff, breaking currently deployed encryption. That is not really a gradual capability that arrives incrementally and gets absorbed slowly into society the way most AI capabilities generally have so far, it is closer to a discrete threshold that, once genuinely crossed, immediately and simultaneously compromises an enormous amount of previously assumed secure infrastructure essentially all at once, all in a single moment rather than gradually over time.

A genuinely useful distinction here separates diffuse ongoing risk from sudden discrete risk. AI risk currently looks more diffuse, ongoing, and continuously distributed across many different domains and many different points in time. Quantum decryption risk looks comparatively more binary and discrete, largely fine right up until a specific threshold is finally crossed, and then potentially catastrophic essentially all at once, in a single sudden moment rather than a gradual accumulation.

Both races share the exact same underlying dangerous incentive structure regardless of which one is ultimately judged worse, first mover advantage strongly rewards speed over caution, and no single actor unilaterally slowing down meaningfully solves the shared, structural problem unless every other competing actor slows down in coordinated tandem as well. That shared structural feature may ultimately matter considerably more, from a policy design standpoint, than which specific race happens to carry the theoretically higher raw peak severity.