# Evaluating the AI 2040 Transparency Plan: The Feasibility of Trustless Compute Verification > An analysis of the proposed Total Research Transparency framework and the geopolitical hurdles of bilateral AGI monitoring. **Published:** July 13, 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:** 1242 **Tags:** AI Governance, Compute Verification, AGI Safety, Geopolitics, Hardware Security **Canonical URL:** https://pseedr.com/risk/evaluating-the-ai-2040-transparency-plan-the-feasibility-of-trustless-compute-ve --- A recent proposal from the AI 2040 project, published on [lessw-blog](https://www.lesswrong.com/posts/NLBmTmryqHGxmWTBR/ai-2040-transparency-plan), advocates for "Total Research Transparency" and a trustless bilateral agreement between the United States and China to verify artificial general intelligence (AGI) development. PSEEDR analyzes the technical and geopolitical feasibility of this framework, specifically evaluating whether hardware-level compute monitoring can realistically bridge the trust gap between rival nations without compromising national security. ## The Core Mechanics of Plan A and Total Transparency The AI 2040 framework, referred to in the source as "Plan A," posits that the primary failure mode in current AI governance stems from poor decision-making by governments and corporations operating in closed-door environments. The prevailing industry norm relies heavily on proprietary development, where frontier models are trained in secret by organizations like OpenAI, Google, and Anthropic. To counter this concentration of power, the authors propose a radical shift toward "Total Research Transparency." Under this proposed model, almost all algorithmic secrets, ongoing experiments, and training runs would be made public immediately. The stated goal is to prevent the abuse of power by ensuring that no single entity can covertly develop AGI capabilities. The source outlines baseline transparency desiderata, which include determining who gets to see algorithmic secrets, who is granted real-time access to data centers, and the latency with which stakeholders are made aware of ongoing training runs. A fundamental pillar of this proposal is the establishment of a trustless verification regime between the United States and China. This regime would require both nations to have direct, verifiable visibility into how compute resources are being utilized, ensuring compliance with potential training slowdown agreements without relying on mutual geopolitical trust. The framework argues that total transparency is the optimal path to improve collective decision-making during the critical phases of AGI development. ## The Technical Feasibility of Trustless Verification The concept of "trustless verification" borrows heavily from cryptographic and decentralized systems, but applying it to physical data centers and sovereign AI infrastructure presents immense technical challenges. The source document outlines the need for real-time data center access and experiment tracking, but it leaves the specific technical mechanisms undefined. From a technical standpoint, implementing such a system would likely require a combination of hardware-level secure enclaves, cryptographic auditing, and tamper-evident compute tracking. To achieve true trustless verification, hardware manufacturers such as NVIDIA or AMD would need to integrate cryptographic signing and auditing capabilities directly into the silicon of their accelerators. This would theoretically allow external auditors to verify the exact nature of the workloads being processed-distinguishing between a massive, coordinated AGI training run and benign, distributed commercial workloads-without necessarily exposing the underlying proprietary data. Trusted Execution Environments (TEEs) and secure boot processes would need to be standardized globally to ensure that the telemetry data reported by a GPU cluster has not been spoofed by the host nation. However, the complexity of modern distributed training, which spans tens of thousands of GPUs across multiple clusters, makes cryptographic auditing highly susceptible to obfuscation. Malicious actors could theoretically partition training runs, mask AGI workloads as standard commercial processing, or utilize undeclared, off-grid compute clusters. Furthermore, verifying the intent of a matrix multiplication operation is inherently difficult; a cluster could be simulating fluid dynamics for aerospace engineering or optimizing the weights of a frontier language model. Developing a verification protocol that is mathematically sound, hardware-enforced, and resilient to state-sponsored evasion remains an unsolved engineering problem. ## Geopolitical Implications of Bilateral Compute Monitoring If the technical hurdles of hardware-level monitoring can be overcome, the geopolitical implications of a US-China compute treaty would fundamentally alter the global technology landscape. The AI 2040 proposal correctly identifies that any effective AGI slowdown agreement cannot rely on trust between geopolitical rivals. A trustless verification regime would require an unprecedented level of mutual intrusion into sovereign digital infrastructure. Historically, arms control treaties relied on physical inspections and satellite telemetry to count warheads or monitor uranium enrichment. Compute, however, is inherently dual-use, highly abstract, and deeply integrated into the modern economy. Granting a foreign adversary real-time visibility into domestic data centers touches the third rail of national security. Intelligence agencies, defense departments, and critical infrastructure providers rely on the same high-performance computing infrastructure that would be subject to this monitoring. The friction in adopting such a framework would be immense, as policymakers must weigh the existential risk of unconstrained AGI development against the immediate national security risk of exposing strategic compute capacity to a rival state. Additionally, major cloud providers would likely resist intrusive monitoring protocols that could compromise the data privacy of their enterprise clients. The proposal highlights a growing faction within the AI safety community that views these radical, hardware-level international treaties as the only viable alternative to the current trajectory, but the political capital required to execute such a treaty is currently non-existent. ## Limitations and Proliferation Risks While the AI 2040 transparency plan offers a theoretical mechanism for bilateral accountability, it introduces severe secondary risks that the source material does not fully resolve. The most glaring limitation of "Total Research Transparency" is the risk of rapid capability proliferation. By making algorithmic secrets and training runs public immediately, the framework inadvertently democratizes access to potentially dangerous AI architectures. The proposal assumes that transparency will improve collective decision-making and prevent power concentration. However, open-sourcing the blueprint for AGI lowers the barrier to entry for malicious non-state actors, rogue states, and highly capable individual developers who are not bound by the US-China bilateral agreement. The source lacks a detailed explanation of how to mitigate this proliferation risk, presenting a paradox where the solution to state-level AGI monopolies directly enables decentralized threat actors. Furthermore, the framework assumes a bipolar AI race, largely ignoring the compute capabilities of the European Union, Middle Eastern sovereign wealth funds, and decentralized open-source communities. A bilateral treaty that successfully throttles US and Chinese AGI development could simply shift the locus of development to jurisdictions outside the monitoring regime. The specific definition and broader scope of "Plan A" within the AI 2040 project remains partially obscured, leaving questions about how this framework would adapt to a multipolar compute landscape. ## Synthesis of the Transparency Paradigm The AI 2040 proposal for Total Research Transparency and trustless verification represents a stark departure from the prevailing norms of corporate secrecy in AI development. By prioritizing public algorithmic sharing and hardware-level monitoring, the framework attempts to solve the governance deficit through radical openness and cryptographic accountability. However, the technical reality of implementing tamper-proof compute verification at a global scale remains highly speculative, requiring unprecedented integration between hardware manufacturers and international regulatory bodies. Furthermore, the tension between preventing state-level AGI monopolies and enabling malicious non-state actors through open-source proliferation remains the central unresolved paradox of the plan. As compute continues to scale, the debate over AI governance will increasingly hinge on whether hardware infrastructure can be effectively regulated without triggering unacceptable national security vulnerabilities. ### Key Takeaways * The AI 2040 Plan A proposes Total Research Transparency to mitigate the risks of closed-door corporate and government AGI development. * A core component of the framework is a trustless bilateral agreement between the US and China to verify compute usage and enforce potential training slowdowns. * Implementing trustless verification would likely require unsolved hardware-level cryptographic auditing and secure enclaves integrated directly into AI accelerators. * The proposal faces immense geopolitical friction, as mutual data center monitoring conflicts directly with national security and sovereign infrastructure protection. * Total transparency introduces severe proliferation risks, potentially democratizing access to dangerous AI capabilities for non-state actors outside the bilateral agreement. --- ## Sources - https://www.lesswrong.com/posts/NLBmTmryqHGxmWTBR/ai-2040-transparency-plan