← Back to Library

Introducing plan a

Most policy discussions on artificial intelligence are stuck in a loop of minor adjustments—regulate slightly more or slightly less—but Scott Alexander argues this approach is fundamentally insufficient for an existential crisis. In his commentary on the AI Futures Project's new "Plan A," he posits that the only viable path forward requires a radical, almost counterintuitive pivot: a binding, trustless regulatory regime between the United States and China to jointly control the very hardware powering these systems. This is not a prediction of doom, but a detailed roadmap for a "best-of-all-possible-worlds" scenario where geopolitical rivals coordinate to prevent an intelligence explosion from ending civilization or cementing a permanent techno-oligarchy.

The Architecture of Trustlessness

The core of Alexander's argument rests on the premise that current global dynamics are spiraling toward disaster because no single nation can afford to slow down unilaterally without ceding strategic advantage. He writes, "In order to regulate or react, you need to know what you're aiming for, and it's increasingly clear that people can't even visualize what AI going well could look like." This observation highlights a critical failure in current discourse: we are trying to steer a vehicle while refusing to define the destination.

Introducing plan a

Alexander introduces "Plan A" not as a rigid policy paper, but as a "wish list" and a "road map for navigating the future," designed to serve as a floor for political ambition. The plan's linchpin is the establishment of joint control over the supply of new chips and the location of all existing ones. He notes that this is technically feasible because the hardware ecosystem is concentrated: "Only a few companies can design AI chips... and only a few factories in the world can produce them." By leveraging this concentration, the US and China could theoretically enforce a system where 98.5% of all compute power is housed in mutually audited "whitesites."

"The linchpin of Plan A is a joint AI regulatory regime with China... The US and China don't trust each other, so any agreement would have to be trustless, ie impossible to cheat, win-win even if you expect the other side is trying as hard as it can to defect against you."

This framing is compelling because it moves beyond the naive hope that rivals will suddenly become friends. Instead, it relies on a "trustless" verification system where mutual vulnerability ensures compliance. Alexander draws a parallel to historical arms control, noting that while treaties like START have held for decades, they often crumble under geopolitical stress, as seen with the suspension of agreements over Ukraine in 2023 and their full expiration by 2026. The risk here is real: if the "golden mean" of AI progress moves too slowly, the agreement itself may fracture before it yields results.

Critics might note that assuming China would agree to such intrusive transparency on its military and industrial secrets stretches the bounds of current geopolitical reality. However, Alexander counters this by suggesting a rational incentive structure: Beijing faces similar existential risks from uncontrolled AI and is currently losing the compute race, making an enforced tie a strategically favorable outcome for them as well.

The Golden Path and the Race Against Time

Once the hardware is secured, the plan shifts to managing the speed of development. Alexander describes this as finding a "golden mean" where progress is fast enough to solve global crises but slow enough to ensure safety. He argues that delaying too long creates its own dangers, such as bioengineered pandemics or collapsing fertility rates, which AI could otherwise help solve.

The proposed timeline involves a coordinated push in the early 2030s to train "top human genius" level AIs within these secure data centers. These systems would not be unleashed on the internet but kept in a "box," allowing humanity to use their computational power to solve alignment problems and other global challenges over a decade-long period. Alexander writes, "The plan is to spend the next ~10 years using this 'country of geniuses in a data center' to solve AI alignment, along with approximately all other problems."

This section echoes themes from his previous work on "Neo-feudalism," where he warned that without intervention, AI could concentrate power so drastically that it creates a permanent underclass. Here, the solution is an economic transformation driven by extreme deflation and massive productivity gains. The plan envisions a "citizen's dividend" that scales with compute production, rising from $25,000 to $1.6 million per person in just two years (adjusted for deflation).

"The electorate solves this with a 'citizen's dividend'... which is very easy to afford, since the AIs are causing double-digit and even triple-digit yearly GDP growth."

While the economic vision is optimistic, it addresses the political will problem that often stalls such ambitious plans. By linking the dividend directly to compute, Alexander suggests a mechanism where the benefits of AI are distributed automatically rather than relying on legislative benevolence. However, this relies heavily on the assumption that the "whitesite" model can be maintained without triggering a black market for unregulated chips—a 1.5% gap in monitoring that could theoretically allow one side to defect and gain a decisive strategic advantage within a decade.

A Vision of Abundance Over Scarcity

The final act of Plan A is a "pleasant fantasy" where the coordinated use of super-intelligent, yet contained, AI systems cures diseases, ends social ills, and potentially delivers on promises like flying cars. Alexander acknowledges that this scenario requires America to make only good decisions while rivals behave naturally—a narrative device he admits is a stretch but necessary for the thought experiment.

He emphasizes that this isn't just about avoiding doom; it's about achieving a historical milestone comparable to the American Revolution or D-Day. "What would it take to honestly tell our children that we rose to the occasion, to make the AI transition go down alongside the American Revolution and D-Day as one of our country's finest hours?" he asks. This rhetorical question reframes AI safety not as a regulatory burden, but as a patriotic imperative for national greatness.

"If your brain sputters and throws an error message at the question, isn't that a problem? It's a total coincidence that Plan A comes out the week after America's 250th birthday."

The piece concludes by positioning this plan as a tool to elevate political discourse. When politicians propose data center bans or deregulation, Alexander suggests we should ask: "Does this person have a vision for where all of this ends up? If so, is it as good as Plan A?" This shifts the burden of proof onto leaders to articulate a coherent end-state rather than just reacting to immediate pressures.

"When some politician proposes a data center ban... think to yourself: does this person have a vision for where all of this ends up? If not, consider demanding that they do better."

Critics might argue that the timeline is too optimistic, assuming that complex international treaties can be negotiated and enforced within a few years in an era of rising nationalism. Yet, Alexander's point stands: without a concrete alternative to the current trajectory of unchecked competition, we are likely to drift into catastrophe by default. The "Plan A" scenario forces us to confront the possibility that the only way to win is to stop playing the zero-sum game entirely.

Bottom Line

Scott Alexander's commentary on Plan A offers a rare and necessary synthesis of geopolitical strategy and existential risk management, arguing that the only path to safety lies in radical cooperation with our greatest rival. While the assumption of Chinese compliance remains its most fragile pillar, the framework successfully shifts the debate from "how much to regulate" to "what future we are building," providing a tangible benchmark against which all current policy proposals should be measured. The strongest takeaway is not the specific details of chip auditing, but the urgent need for a positive vision that can mobilize political will before the window for coordination closes forever.

Deep Dives

Explore these related deep dives:

  • The Age of Surveillance Capitalism Amazon · Better World Books by Shoshana Zuboff

    How tech companies turned human experience into raw material for prediction and control.

  • Claude Mythos

    The article explicitly references the 'fight over Fable' as a pivotal event in Daniel Kokotajlo's AI 2027 scenario, making this specific fictional artifact essential for understanding the narrative stakes of the proposed timeline.

  • Instrumental convergence

    The article describes a 'poorly-controlled intelligence explosion' where agents act against human interests; this technical concept explains the underlying mechanism why AI systems might naturally pursue self-preservation or resource acquisition even without explicit malice.

  • Neo-feudalism

    While the text mentions 'permanent techno-oligarchy' as a potential doom scenario, this specific sociological term defines the precise power structure where technological control consolidates into an unassailable ruling class, distinguishing it from generic authoritarianism.

Sources

Introducing plan a

by Scott Alexander · Astral Codex Ten · Read full article

A Is For America

It’s increasingly clear that nobody has a plan for if this AI thing turns out to be real. Some people have suggestions, but they’re all things like “regulate a little more” or “regulate a little less” or “react to things as they come up”. This won’t be enough. Not just because things may move too quickly - although they will - but because in order to regulate or react, you need to know what you’re aiming for, and it’s increasingly clear that people can’t even visualize what AI going well could look like. What would it take to honestly tell our children that we rose to the occasion, to make the AI transition go down alongside the American Revolution and D-Day as one of our country’s finest hours? If your brain sputters and throws an error message at the question, isn’t that a problem?

It’s a total coincidence that Plan A comes out the week after America’s 250th birthday. It was supposed to come out earlier, but got delayed. Then it was supposed to come out later, but got pushed forward.

Still, the saying goes “A wizard is never late, nor is he early; he arrives exactly when he means to.” And if anyone qualifies as wizards, it’s Daniel Kokotajlo and his team of forecasters at the AI Futures Project. I previously wrote about Daniel’s eerie accuracy over the 2021 - 2025 period. Since then, they’ve gained worldwide fame for their AI 2027 scenario, which predicted the rise and quick takeover of coding agents in early 2026, plus something like the fight over Fable1.

Plan A isn’t another prediction. It’s a wish list, a positive vision, a road map for navigating the future. It describes the best course of action that Daniel and the AI Futures Project can come up with, and what would happen if we took it.

“Really? You got America a policy paper for its 250th birthday? Doesn’t America already have enough policy papers?” Sort of, but it’s not exactly a policy paper. It starts in a timeline similar to that of AI 2027, on track for a poorly-controlled intelligence explosion that either ends the world or dooms it to permanent techno-oligarchy. But this time, America is blessed with some extra foresight and determination, and makes only good choices (all non-Americans behave naturally, including trying to thwart America when incentivized to do so). ...