# Analyzing 'Plan A': The Emerging AI Governance Blueprint from Top Forecasters > A proposed strategic framework attempts to bridge the gap between AI accelerationism and total shutdown through proactive governance and compute controls. **Published:** July 11, 2026 **Author:** PSEEDR Editorial **Category:** risk **Content tier:** free **Accessible for free:** true **Editorial format:** analysis **News quality eligible:** true **Source count:** 1 **Word count:** 1093 **Tags:** AI Governance, Compute Controls, AI Safety, Forecasting, Technology Policy **Canonical URL:** https://pseedr.com/risk/analyzing-plan-a-the-emerging-ai-governance-blueprint-from-top-forecasters --- As artificial intelligence capabilities accelerate toward predicted 2026-2027 milestones, the AI safety community is transitioning from passive forecasting to active policy formulation. A recent analysis on [lessw-blog](https://www.lesswrong.com/posts/z9tXCGogEgkgHSh8G/introduction-for-and-reactions-to-plan-a) examines "Plan A," a strategic framework proposed by prominent AI forecasters that attempts to navigate the transition to superintelligence. PSEEDR analyzes this blueprint through the lens of pragmatic AI governance, evaluating whether a forecaster-led vision can realistically bridge the divide between unregulated accelerationism and total shutdown scenarios. ## The Shift from Forecasting to Policy Formulation The introduction of Plan A marks a notable maturation in the AI safety discourse. Historically, much of the community's output has focused on timeline forecasting and risk identification. The authors of Plan A, Daniel Kokotajlo and Ryan Greenblatt, have established significant credibility in these areas. Kokotajlo authored the highly scrutinized "AI 2027" and "What 2026 Looks Like" predictions, while Greenblatt, Chief Scientist at Redwood Research, ranked as the second most accurate AI forecaster in 2025 out of over 400 participants. Their transition from predicting the future to proposing a structured roadmap to manage it indicates a recognition that passive observation is no longer sufficient. By leveraging their forecasting track records, the authors are attempting to establish a credible foundation for proactive governance. The lessw-blog analysis notes that while past predictive success does not guarantee the viability of their policy proposals, it does warrant serious consideration from the broader technical and regulatory communities. Plan A is positioned not as a finalized legislative draft, but as a forcing function to shift the Overton window, moving the conversation toward actionable, intermediate steps rather than abstract existential dread. ## Defining the Strategic Alternatives: Plan A vs. Plan S The core of the lessw-blog discussion contrasts Plan A with alternative strategies, most notably "Plan S," which stands for Shutdown. Plan S represents the extreme end of the AI safety spectrum, advocating for a hard halt to advanced AI development to prevent potential catastrophic outcomes. However, a global shutdown faces insurmountable coordination hurdles, requiring unprecedented international cooperation and the likely use of extreme state power to enforce compliance. Plan A operates on a principle of "selective optimism." It acknowledges the rapid diffusion of AI technologies and the massive economic growth incentives driving the industry, but proposes that this trajectory can be steered rather than stopped. The framework attempts to address the geopolitical realities of AI development, including the race conditions between nation-states and the complexities of maintaining technological competitiveness. By offering a positive vision that allows for continued, albeit heavily managed, technological progress, Plan A aims to build a broader coalition of support than the politically unviable Plan S. ## The Compute Control Controversy A central pillar of Plan A's governance strategy relies on compute controls, monitoring and restricting the physical hardware required to train and deploy frontier AI models. Because software and algorithms are non-rivalrous and easily proliferated, physical compute infrastructure remains the most viable chokepoint for regulatory enforcement. The lessw-blog analysis highlights that this is the most contentious aspect of the framework. Critics argue that implementing stringent controls over compute resources equates to an authoritarian dystopia, requiring pervasive surveillance of data centers and supply chains. Conversely, proponents view compute governance as the only pragmatic mechanism to prevent unregulated superintelligence. The debate frequently hinges on underlying assumptions about future AI capabilities; as cited in the source, figures like Vitalik Buterin argue that the crux of the disagreement often lies in differing projections of AI's ultimate limits and the speed at which those limits will be reached. If superintelligence is imminent, drastic compute controls appear necessary; if it is distant, they appear draconian. ## Implications for Global AI Governance If the principles outlined in Plan A gain traction among policymakers, the implications for the technology ecosystem will be profound. The burden of regulatory compliance will shift heavily onto hardware manufacturers, semiconductor foundries, and cloud hyperscalers. Companies operating massive data centers would effectively become the enforcement arms of global AI governance, required to implement strict verification protocols for compute usage. Furthermore, Plan A's approach requires a fundamental restructuring of international technology agreements. To prevent race conditions, compute controls cannot be unilateral; they require a multilateral consensus that binds competing superpowers. This necessitates a shift from traditional arms control frameworks, which monitor physical weapons, to a new paradigm of computational arms control. The friction in adopting such a framework will be immense, as nations must balance the economic imperative of AI supremacy against the shared risks of unaligned superintelligence. ## Limitations and Open Questions Despite its ambitious scope, the current iteration of Plan A, as analyzed in the source material, leaves several critical gaps. The exact technical and policy mechanisms required to enforce global compute controls are not fully detailed. It remains unclear how regulators can monitor distributed training runs or prevent the smuggling of advanced chips without establishing the very surveillance apparatus that critics fear. Additionally, the specific operationalization of alternative strategies like Plan S is left largely unexplained, making it difficult to rigorously benchmark Plan A's comparative advantages. The framework also assumes a degree of rational state actor behavior that may not hold true in a high-stakes geopolitical race. Until the proponents of Plan A can provide concrete, enforceable mechanisms that navigate the tension between competing technological ambitions, the blueprint remains a theoretical exercise rather than a deployable policy. The emergence of Plan A represents a critical pivot in the AI safety ecosystem, signaling a move from theoretical risk modeling to pragmatic policy design. By proposing a managed, compute-centric approach to AI governance, top forecasters are attempting to chart a middle path between the economic momentum of accelerationism and the paralyzing constraints of total shutdown. While the framework currently lacks the granular enforcement mechanisms necessary for global implementation, it successfully forces the industry to confront the physical realities of AI regulation. As the timeline to advanced AI compresses, frameworks like Plan A will increasingly define the parameters of the regulatory debate, shifting the focus from whether AI should be governed to how its underlying infrastructure can be controlled. ### Key Takeaways * Plan A is a proactive AI governance framework proposed by top forecasters, marking a shift from passive timeline prediction to active policy formulation. * The framework relies heavily on compute controls as a regulatory chokepoint, a strategy that critics argue risks authoritarian surveillance but proponents view as necessary. * Plan A attempts to offer a pragmatic middle ground between unregulated AI acceleration and 'Plan S' (total shutdown), acknowledging the realities of economic growth and geopolitical race conditions. * Significant limitations remain regarding the exact technical mechanisms for enforcing global compute controls without triggering international conflict. --- ## Sources - https://www.lesswrong.com/posts/z9tXCGogEgkgHSh8G/introduction-for-and-reactions-to-plan-a