The current atmosphere in the AI industry feels less like a scientific pursuit and more like a high-stakes arms race. In the boardrooms of Silicon Valley and the research hubs of Beijing, there is a pervasive, unspoken belief that the first entity to achieve superintelligence—an AI that surpasses human capability across every single domain—will effectively own the future. This winner-takes-all mentality has triggered a cycle of reckless acceleration, where safety benchmarks are treated as hurdles to be cleared quickly rather than safeguards to be respected. The pressure to ship the next frontier model is no longer just about market share; it is about the perceived necessity of achieving dominance before a competitor does.
The Architecture of Plan A
To counter this trajectory, a strategic framework known as Plan A has been proposed. This is not a mere set of ethical guidelines, but a rigorous international agreement designed to intentionally decelerate the transition to superintelligence, pushing the target date back to 2040. The core philosophy of Plan A is the replacement of secret, competitive development with a transparent, collaborative front. By opening up research and ensuring that multiple nations and corporations participate in the development process, the plan seeks to prevent any single group from monopolizing the technology.
The timeline for Plan A is precise and phased. It begins in 2029, with a critical diplomatic agreement between the United States and China to avoid a reckless race toward superintelligence. This geopolitical truce is the foundation upon which the rest of the schedule rests. By 2030, the industry will likely reach a tipping point where AI can fully automate its own research and development. In a competitive market, this would create a recursive loop of self-improvement, potentially leading to superintelligence by the end of 2030. Plan A explicitly mandates that this automation loop be avoided. Instead, from 2030 to 2035, AI capabilities will be expanded only to the level of the highest-performing human experts, ensuring the technology remains within the boundaries of human comprehension and management.
In 2035, the plan calls for a total temporary suspension of development. This pause is designed to ensure that human control mechanisms are fully matured and locked in before the final leap. Only in 2040 will this suspension be lifted, allowing for the controlled transition to superintelligence. To make this timeline viable, Plan A suggests immediate interim measures. These include strict limits on the compute budgets allocated to AI R&D and policies to narrow the performance gap between models used internally by companies and those released to the public. Furthermore, it recommends that governments take an active role by securing AI talent, monitoring chip supply chains, and enforcing strict export controls to prevent clandestine acceleration.
The Power Paradox and Compute Destruction
While much of the public discourse around AI focuses on alignment—the technical challenge of ensuring an AI does what we want—Plan A identifies a more immediate and visceral danger: the concentration of power. The realization here is that even a perfectly aligned AI is a weapon if it is owned by a tiny elite. If a single individual or a small group of people holds a monopoly over a superintelligent system for even a few months, they possess the capability to reshape the global order, effectively seizing control of the planet. The risk is not just that the AI might go rogue, but that the human owner might use a perfectly functioning AI to establish an unbreakable hegemony.
This is why Plan A rejects alternative scenarios like Plan C, which suggests only a brief slowdown of a few months, or Plan D, which accepts the immediate race. Such plans are viewed as insufficient because they do not address the inherent instability of a power imbalance. When one nation or company achieves a decisive lead in superintelligence, it creates an existential crisis for everyone else, maximizing global anxiety and significantly increasing the probability of a preemptive world war. The speed of development is therefore not a technical variable, but a geopolitical trigger.
To prevent this, Plan A introduces a concept akin to the Cold War's Mutual Assured Destruction, termed Mutual Assured Compute Destruction. In this system, the global community establishes a framework where the destruction of one party's compute resources—the semiconductors and server infrastructure required for AI training—would trigger the simultaneous destruction of the aggressor's own resources. By creating a state of mutual vulnerability, the incentive to secretly surge ahead is neutralized. This system relies on total transparency and verifiable safety mechanisms, shifting the industry from a closed-door race to a monitored, multi-polar expansion. Instead of a few companies racing in secret, dozens of firms across multiple nations would expand their capabilities in lockstep, ensuring that no single entity ever gains a decisive, monopolistic advantage.
The transition to superintelligence is an event that cannot be undone. By treating the arrival of ASI as a coordinated global event rather than a corporate milestone, Plan A attempts to decouple technological progress from geopolitical instability. The goal is to ensure that when the threshold is finally crossed in 2040, the world is governed by a system of shared control rather than the whims of a single winner.
This agreement is not a concession of technical ambition, but a calculated survival strategy to prevent the ultimate monopoly.
