How to Secure Advanced U.S. Chips
A policy memo arguing the U.S. should close gaps in advanced chip export controls, covering remote cloud access, smuggling, H200 exports and chipmaking equipment, to limit China's frontier AI development.
Research / Topics
Compute & hardware
Chips, export controls, compute thresholds, data centres and energy, hardware-enabled mechanisms.
Focus All
Source All
Date Any time
Explaining “AI might obliterate you” to senior policymakers in India
A former DeepMind safety researcher transcribes a talk urging Indian policymakers to take loss-of-control risks from rogue AI agents seriously, and reflects on its rhetoric.
Cloud Compute: A Policy Escalation Framework
Argues that Washington needs a strategy for Chinese developers renting frontier U.S. AI compute through overseas cloud providers, proposing a policy escalation framework to address national security risks.
Detecting AI Training in Datacentre Power
Proposes detecting large AI training runs from datacentre power draw via Viterbi tracking of iteration cadence, showing it is robust to most operator hiding strategies and useful for compute verification under AI governance agreements.
Quick Thoughts on Overhang Arguments
A rough critique of 'overhang' arguments in AI safety, which justify accelerating progress now to slow later risk, using analogies from other fields and arguing such reasoning is suspect and prone to motivated reasoning.
How far is ZKP from compute verification in high-stakes environments?
Estimates how far zero-knowledge proofs of LLM inference are from cost-competitive with auditing for verifying AI compute in international agreements, measuring roughly 1,700-3,800x proving overhead and hash-function compatibility issues.
Proof of useful work for verifying AI treaties
Literature review of proof of useful work, a cryptographic method for GPUs to prove compute use, as a way to verify international AI pause treaties, finding open gaps in guarantees and hardware assumptions.
How to Make an AI Deal With China: Trade Throttling for Pacing
Argues that the US should use the threat of throttling AI chip or development trade to secure a mutual US-China deal to pace frontier AI development, citing recent agent containment incidents.
Can Power Draw Constrain Covert Compute? Limits of Analogue Verification for AI Governance
Derives a bound on how much hidden computation power-draw measurements can exclude for frontier AI compute agreements, finding analogue channels alone weak but useful when combined with re-execution.
Skeuomorphic AI Safety
Proposes making AI hardware resemble bounded humans, such as fused-weight ASICs, so existing institutions stay workable, arguing this is more robust than alignment-only approaches.
Plan R: AI Safety by ASICs
Proposes separating frontier AI training from deployment and enforcing the split with hardware-level ASIC inference, latency limits, and anti-dogfooding rules to reduce catastrophic AI risk.
Against Export Controls (and China Threat Models)
Argues against U.S. export controls on AI chips as a misalignment-reducing policy, contending they may speed U.S. recursive self-improvement and hinder a bilateral treaty with China.
Weight smuggling likely defeats attempts to cap FLOPs per training run
Argues that per-run FLOP caps for frontier training can be covertly bypassed by smuggling weights into new runs or aggregating parallel runs, and suggests a global training budget instead.
Reducing the Resource Gap Between Lab and External Safety Researchers
Geodesic Research describes compute and model-access bottlenecks that independent AI safety organisations face versus frontier labs, and proposes philanthropic interventions such as shared compute pools and brokered agent access.
Could Underwater Data Centers Pose a Risk to AI Treaty Verification?
Assesses whether underwater data centers could enable covert frontier AI training that evades treaty verification, finding them feasible only for well-resourced states and detectable via construction signatures and surveillance.
Who computes?
Formalises AI doom versus alignment deliverance as a competing-risks survival model driven by compute allocation and pacing, examining how response-function shapes change whether speed or split matters.
Priorities for Frontier AI Policy
A policy agenda for frontier AI, proposing lifecycle oversight including internal models, American AI leadership measures, and national resilience against AI-enabled cyber, bio and distillation threats.
Recommendations for People Getting into Technical AI Governance Research
MIRI shares reading lists and research advice for newcomers to technical AI governance, drawn from its 2026 fellowship, to help researchers enter the field.
Beyond Training: A Feasibility Taxonomy for Inference-Time AI Governance
Proposes a feasibility taxonomy of twenty inference-time governance mechanisms, rated for readiness and stress-tested against adversary models and governance scenarios, finding no mechanism adequate against high-capability state-level deployers.
The Scramble: getting in position to pace the frontier
Proposes a rapid crisis playbook for pacing frontier AI before recursive self-improvement, covering an interim US-China deal, a voluntary company charter, verification with existing tools, and prep work.
Near-Term Verification Methods for AI Chip Exports
Proposes near-term verification mechanisms for AI chip export controls, covering chip location, end-user legitimacy and prohibited end-use, implementable within the US BIS framework using private-sector auditors.
Also: IAPS research
AI breakout time: a mechanism for an enforceable AI pause
Proposes designing an enforceable AI pause around a breakout-time threshold, arguing it should be set where rogue actors would need years to reach dangerous capabilities, with adaptive adjustment and supply-chain controls.
Open Compute Verification Observatory as a First Metrological Layer for Technical AI Governance
Proposes a shared, public-cluster observatory of GPU telemetry and other physical signals to verify AI training runs, as a first measurement layer for frontier AI governance.
Workload Identification with Physical Side Channels for AI Governance
Shows an external power-draw side channel can identify AI training versus inference on an H200 GPU for compute governance, and a hardened detector still catches most evasion attempts.
How the American Executive Could Control AI Companies
Analyses US federal statutes and executive powers that let the president unilaterally influence or control frontier AI companies, and the limits imposed by Congress and courts.
How to pace the US frontier
Proposes concrete domestic US policies to pace frontier AI development, including compute allocation minimums for safety research, caps on AI R&D use, and safety-case-based existential risk thresholds enforced by third-party assessors.
Traffic Shaping for Workload Classification
Proposes a network-based verification design, traffic shaping plus randomized routing, that estimates it would make covert frontier training runs economically infeasible and so support treaty compliance.
Five counterintuitive insights from Plan A
Lays out five counterintuitive takeaways from the Plan A AI slowdown proposal, covering safety effort versus slowdown time, transparency, compute arms control, deal-breakdown risk, and the security-transparency tradeoff.
An International AI Slowdown Is Ready Whenever Politicians Are
Argues that an international AI slowdown can be verified with human inspectors given whole-lab access to frontier labs, and that the technical verification barrier is not a reason to delay it.
My Assessment of Plan A's Compute Verification Strategy (+ open questions)
Analyses technical mechanisms for verifying that datacenter compute is used only for inference, such as interconnect limits, memory wipes, replay, and zero-knowledge proofs, to slow unmonitored frontier training.
Is a pause enforceable? New paper out!
Proposes a taxonomy for verifying and enforcing international AI pauses via compute monitoring, estimating that pauses are enforceable for training runs above roughly 10k H100-equivalents absent concealment.
Plan A, by AI-2040
A commentary on the AI-2040 'Plan A' scenario, a proposed international treaty to slow AI progress, assessing its governance, enforcement, and alignment assumptions, including reliance on AI lie detection.
Estimating LLM Training FLOPs on the Nvidia Jetson Orin Nano
Proposes and tests a side-channel method that estimates LLM training FLOPs from Jetson GPU power and memory readings, to help verify compute thresholds in AI treaties, reaching about 4.5% mean error.
All Watched Over
An essay arguing that AI power and control should be decentralized rather than concentrated in a few firms, governments, or the AI itself, and that democratic checks must be built into how AI is deployed.
Remote Access Security Act (RASA)
Proposes closing a gap in US export controls so advanced AI chips rented through cloud services remain covered, and recommends broadening the Senate bill's definition of cloud infrastructure.
A short summary of AI 2040: Plan A
Summarises AI 2040: Plan A, an international AI-race slowdown treaty with compute verification and transparency that still funds alignment research, and lists its failure modes.
5 "Plan A" scenarios
Brainstorms governance options for advanced AI development, including chip-level monitoring, joint projects, deterrence, capability red lines, and liability regimes, and how they could combine into a stronger strategy than Plan A.
Plan A: Suggestions For Further Work
A research agenda from the AI Futures Project listing open questions on Plan A for navigating the intelligence explosion, covering alternative scenarios, covert projects, US domestic governance, verification, economics and geopolitics.
Plan A's problem with dry tinder
Critiques the AI Futures Project's Plan A pause-and-compute strategy, arguing that accumulated compute and software overhang (dry tinder) could make a deal breakdown far more dangerous, and that mutually assured compute destruction is fragile.
Can the U.S. and China Deny AI?
A quantitative model of how destroying US and Chinese AI compute would delay superintelligence development, finding strikes cause only 1-5 year delays that nationalisation can offset, undermining kinetic deterrence.