REVIEW 3 major objections 5 minor 68 references
Domestic frontier AI regulation, an IAEA for AI, an NPT for AI, and a US-led Allied Public-Private Partnership for AI: Four institutions for governing and developing frontier AI
T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash
Pith's one-line read This paper argues that frontier AI governance can be built on compute governance: controlling the chips and training runs that define frontier models gives governments a physical, auditable lever, and the paper designs four institutions…
desk verdict A well-scoped policy exploration of four compute-anchored international AI institutions; the pieces aren't new but the integrated package and phased roadmaps are, and the main soft spot is an unexamined verification threat model. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central mechanism is compute-indexed governance: regulatory scrutiny attached to the FLOP of a training run, with a threshold (around $10^{26}$ FLOP today) above which pre-notification, risk assessment, security standards, usage reporting, and release gates apply. The enforcement lever is the concentrated supply chain for state-of-the-art chips—more concentrated than the fissile-material supply chain, with one to three companies at many steps. The paper proposes tracking chips through unique IDs and registries, chain-of-custody audits, and on-chip hardware features (delay-based location verification, network limits, remote enforcement, multiparty control), and using access to chips as the carrot and stick: compliant states are certified by an international AI agency and may import advanced chips; non-compliant states may not.
What would settle it
Track the compute-to-capability ratio over the next frontier model generations: if a generation arrives whose capabilities advance at the usual frontier rate while training compute grows by less than, say, a factor of 10 instead of the projected 1000x, or if an equivalent frontier model is trained across many small distributed clusters using ordinary hardware, the assumption that frontier runs are huge and concentrated fails and the proposed institutions lose their enforcement handle.
Extended reading notes
Core claim
The central claim is that compute governance can underpin international institutions for frontier AI. The author assumes scaling laws continue: frontier progress is substantially driven by increases in training compute, so frontier training runs remain huge, visible, expensive, and concentrated in a handful of actors and chip producers. On that basis, the paper argues that chip supply can serve as the enforcement lever: states without a certified domestic frontier-regulation regime would not receive advanced chips, and states that join the regime gain access. The four institutions reinforce each other—domestic regulation creates standards, an international agency develops and monitors safeguards, a chips agreement withholds compute from non-compliant states, and a joint allied megaproject consolidates the riskiest training runs in a secure, legitimate setting—so that no insecure, unsafe, or secret frontier training run happens without knowledge and oversight.
Load-bearing premise
That frontier AI progress will keep being driven mainly by ever-larger training runs, so the biggest models keep needing huge, visible, expensive concentrations of top-of-the-line chips that a government can count and control.
Editorial extensions
If this is right
- Domestic frontier AI regulation can move from notification to pre-approval: structured risk assessments before training, security standards, data centre usage reports and audits, and release gates before deployment.
- An International AI Agency could grow out of existing national safety-institute cooperation and AI summits, first running evaluations, then harmonizing standards, then monitoring compliance, eventually auditing data centres.
- A Secure Chips Agreement could extend today's export controls into a verified non-proliferation regime: unique chip IDs, registry, chain of custody, random audits, and on-chip security features, with non-signatories cut off from advanced chips.
- A US-led allied public-private partnership could pool the hundreds of billions of dollars that frontier clusters may soon require, and centralize the riskiest training runs under higher security and legitimacy, with access to models distributed through licensed entities.
- Because frontier chips quickly become obsolete and must be continually repurchased, withholding ongoing access keeps non-participants from competing at the frontier, making participation in the regime self-reinforcing.
Reading between the lines
- Editorial inference: if the compute-based architecture works, the same physical audit trail could later support other international bargains—such as verified limits on AI inference for military purposes or verified benefit-sharing from public-goods models—because the hard part, counting and tracking the hardware, is already in place.
- Editorial inference: the logic also predicts pressure toward consolidation even without a formal partnership: if costs reach the $10B–$100B range, compute-sharing consortia or state-backed clusters among allied states become increasingly likely, and some may form before a formal US-led project does.
- Editorial inference: a testable consequence is that the governance value of chip tracking rises with supply-chain concentration; if new chip fabs or distributed low-power training architectures break the current chokepoints, the institutions would need to shift their monitoring from training runs to model deployment, which is technically harder.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This paper argues that compute governance can serve as the backbone of four complementary institutions for frontier AI: domestic compute-indexed regulation, an International AI Agency modeled on the IAEA, a Secure Chips Agreement modeled on the NPT, and a US-led Allied Public-Private Partnership for frontier AI. It develops each as a phased roadmap, discusses incentives such as access to state-of-the-art chips, and stresses that the whole package depends on continued scaling of training compute. The paper is explicitly exploratory: it repeatedly flags downside risks, names open questions, and does not claim to settle the policy debates it surveys.
Significance. The paper's strength is its comprehensive and candid integration of compute governance into concrete institutional design. It goes beyond slogan-level 'IAEA for AI' proposals by specifying phases, actors, and incentive mechanisms, and it engages the existing critical literature on the IAEA analogy. The author is unusually transparent about limitations, including the dependence on scaling laws, the likelihood of an allied-only rather than truly global regime, and the risk of concentrating power in a single megaproject. If the compute-governance handle holds, this would be a useful blueprint for international AI governance; the explicit caveats make the conditional nature of the claim easy to test. The paper does not provide formal proofs or machine-checked artifacts, but it does provide a systematic policy argument with references to the relevant technical literature.
major comments (3)
- [§1 and §2.4/§4.2–4.4] The central argument requires not merely that frontier training runs use large compute, but that this compute is externally verifiable. The paper's monitoring chain—cryptographic challenges, chip registries, delay-based geolocation, secure enclaves, and remote attestation—is described with 'could' proposals and citations to early-stage work, but no adversarial threat model is supplied. A proof-of-work challenge confirms that a chip is busy, not what computation it performs; delay-based geolocation can be proxied; secure enclaves and remote attestation have a history of side-channel and firmware attacks; and chain-of-custody audits depend on the same firms being audited. The paper needs either a quantitative bound on the diversion rate and evasion effort needed to assemble a 1e26-FLOP cluster, or an explicit statement that the effectiveness of these monitoring mechanisms is an unresolved empirical question that the institutional design cannot yet assume. As written, the assurance that governments would gain from the IAIA and Secure Chips Agreement is asserted rather than demonstrated.
- [§5.1 and Table 5] The claimed advantages of a US-led Allied PPP over private and national projects rest on a seven-row ordinal ranking with no justification of the criteria weights, no sensitivity analysis, and no discussion of how the rankings were derived. For example, the paper asserts without evidence that a US-led Allied project would be 'safer' and 'less prone to misuse' than a private project, while the preceding paragraph notes that companies face strong competitive pressures; the opposite conclusion is at least arguable. The table should be reframed as an illustrative summary of the author's qualitative views, or supported by a structured comparison with explicit evidence for each row.
- [§4.5 and §5.8] The proposed non-proliferation regime is explicitly limited to US allies and excludes China, but the paper does not fully address how an 'International' AI Agency that excludes the world's second-largest AI developer can still perform the IAEA-like function of providing global assurance. The paper acknowledges this tension in §3.7, but the four institutions are then de facto an allied bloc; the conclusion's claim that governments can be 'reassured' about global frontier AI development overstates what an allied-only regime can achieve. This is a scope limitation that should be stated as such in the conclusion, since it affects the paper's central claim about international reassurance.
minor comments (5)
- [§4, Table 2 caption] The caption begins 'F or both nuclear and frontier AI' and should read 'For both nuclear and frontier AI'.
- [§5, Table 6] The table contains 'succesful' and 'particpating'; these should be 'successful' and 'participating'.
- [Figure 5 caption] The caption 'A WS’ nuclear-powered 960-MW data centre campus' should read 'AWS's nuclear-powered...'.
- [Introduction and Appendix] Compute notation is inconsistent: the text uses '10 26 FLOP', '1021 FLOP', '10 29 FLOP', and similar forms without consistent superscript formatting. Please unify these to a single notation such as 1e26 FLOP.
- [§3.7] The discussion of whether the IAIA should include China is useful, but the paper could more explicitly define 'international' when the likely regime is allied-only, so that readers are not misled by the title's promise of an international agency.
Circularity Check
No significant circularity: the institutional proposals are argued from a stated scaling assumption and independently sourced evidence, not from their own conclusions.
full rationale
This is an exploratory policy paper, not a derivation or empirical prediction. The central claim that compute governance can underpin four frontier-AI institutions rests on the explicitly stated assumption that scaling laws continue and that frontier training runs remain large, visible, and concentrated. The author acknowledges that if this assumption fails, the scenarios become much less motivating, so the claim is not smuggled in as a conclusion. The paper's factual inputs, such as supply-chain concentration, chip-registry proposals, proof-of-work monitoring, and cost trends, are cited to external reports or to the author's prior collaborative work (Sastry et al. 2024), but these are background evidence that is externally falsifiable and independently corroborated by other cited sources (e.g., Shavit 2023, Aarne et al. 2024, Cheng 2024). The monitoring mechanisms are presented as 'could' proposals rather than established facts, and the institutional package does not assume its own effectiveness to prove itself. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors' prior work, and the paper explicitly disclaims being the first to propose these options. The only mild self-citation is the use of Sastry et al. for monitoring and registry ideas, but it is not load-bearing. Thus there is no circular derivation.
Assumptions & free parameters
assumptions (4)
- domain assumption Frontier AI progress will continue to be substantially driven by increases in training compute, and 'scaling laws' will continue for the largest models.
- domain assumption Advanced AI chip supply chains are and will remain concentrated in a few US-allied companies and countries.
- domain assumption Frontier AI poses significant national security risks that justify preemptive governance.
- domain assumption The IAEA and NPT governance model is sufficiently applicable to frontier AI to serve as a template.
invented entities (3)
-
International AI Agency (IAIA)
-
Secure Chips Agreement
-
US-led Allied Public-Private Partnership for Frontier AI
Cite this review
Pith. "Pith review of Domestic frontier AI regulation, an IAEA for AI, an NPT for AI, and a US-led Allied Public-Private Partnership for AI: Four institutions for governing and developing frontier AI." pith.science (2026). https://pith.science/paper/4VJYZETY
@misc{pith2026250706379,
author = {Pith},
title = {Pith review of: Domestic frontier AI regulation, an IAEA for AI, an NPT for AI, and a US-led Allied Public-Private Partnership for AI: Four institutions for governing and developing frontier AI},
year = {2026},
howpublished = {\url{https://pith.science/paper/4VJYZETY}},
note = {Machine review of arXiv:2507.06379}
}
read the original abstract
Compute governance can underpin international institutions for the governance of frontier AI. To demonstrate this I explore four institutions for governing and developing frontier AI. Next steps for compute-indexed domestic frontier AI regulation could include risk assessments and pre-approvals, data centre usage reports, and release gate regulation. Domestic regimes could be harmonized and monitored through an International AI Agency - an International Atomic Energy Agency (IAEA) for AI. This could be backed up by a Secure Chips Agreement - a Non-Proliferation Treaty (NPT) for AI. This would be a non-proliferation regime for advanced chips, building on the chip export controls - states that do not have an IAIA-certified frontier regulation regime would not be allowed to import advanced chips. Frontier training runs could be carried out by a megaproject between the USA and its allies - a US-led Allied Public-Private Partnership for frontier AI. As a project to develop advanced AI, this could have significant advantages over alternatives led by Big Tech or particular states: it could be more legitimate, secure, safe, non-adversarial, peaceful, and less prone to misuse. For each of these four scenarios, a key incentive for participation is access to the advanced AI chips that are necessary for frontier training runs and large-scale inference. Together, they can create a situation in which governments can be reassured that frontier AI is developed and deployed in a secure manner with misuse minimised and benefits widely shared. Building these institutions may take years or decades, but progress is incremental and evolutionary and the first steps have already been taken.
Figures
Figures from the paper (3 more)
Reference graph
Works this paper leans on
-
[1]
Aarne, O., Fist, T. and Withers, C. (2024) Secure, Governable Chips: Using On-Chip Mecha- nisms to Manage National Security Risks from AI & Advanced Computing. CNAS. Available at: https://www.cnas.org/publications/reports/secure-governable-chips
work page 2024
-
[2]
Abungu, C., Malonza, M. and Adan, S.N. (2023) ‘Can apparent bystanders distinctively shape an outcome? Global south countries and global catastrophic risk-focused governance of artificial intelligence’. arXiv
work page 2023
-
[3]
Afina, Y. and Lewis, P. (2023) ‘The nuclear governance model won’t work for AI’, Chatham House – International Affairs Think Tank, 28 June. Available at: https://www.chathamhouse.org/2023/06/nuclear- governance-model-wont-work-ai (Accessed: 17 April 2024)
work page 2023
-
[4]
Anderljung, M. et al. (2023) ‘Frontier AI Regulation: Managing Emerging Risks to Public Safety’. arXiv
work page 2023
-
[5]
(2024) Situational Awareness: The Decade Ahead
Aschenbrenner, L. (2024) Situational Awareness: The Decade Ahead. San Francisco, CA: Situ- ational Awareness
work page 2024
-
[6]
Avin, S. et al. (2021) ‘Filling gaps in trustworthy development of AI’, Science, 374(6573), pp. 1327–1329
work page 2021
-
[7]
Baker, J.N. (2024) ‘International law and advanced AI: exploring the levers for “hard” control’, Institute for Law & AI, 20 July. Available at: https://law-ai.org/international-law-and-advanced-ai/ (Accessed: 12 August 2024)
work page 2024
-
[8]
(2013) ‘Subatomic Politics: Particle Physics and Science Diplomacy’, World Politics Review, 19 March
Barish, B.C. (2013) ‘Subatomic Politics: Particle Physics and Science Diplomacy’, World Politics Review, 19 March
work page 2013
Show all 68 references
-
[9]
& Ruhl, C
Belfield, H. & Ruhl, C. (2022) ‘Why policy makers should beware claims of new ‘arms races’, Bulletin of the Atomic Scientists, 14 July
2022
-
[10]
(2022) ‘The Rival AI Deployment Problem: a Pre-deployment Agreement as the least-bad response’, EA Forum, 23 September
Belfield, H. (2022) ‘The Rival AI Deployment Problem: a Pre-deployment Agreement as the least-bad response’, EA Forum, 23 September
2022
-
[11]
(2023) Great British Cloud and BritGPT: the UK’s AI Industrial Strategy Must Play to Our Strengths
Belfield, H. (2023) Great British Cloud and BritGPT: the UK’s AI Industrial Strategy Must Play to Our Strengths. Labour for the Long Term. Available at: https://www.labourlongterm.org/briefings/great- british-cloud-and-britgpt-the-uks-ai-industrial-strategy-must-play-to-our-st...
2023
-
[12]
Bengio, Y. et al. (2024) International Scientific Report on the Safety of Advanced AI - Interim Re- port. UK & Korean Governments. Available at: https://www.gov.uk/government/publications/international- scientific-report-on-the-safety-of-advanced-ai
2024
-
[13]
(2024) Location Verification for AI Chips
Brass, A. (2024) Location Verification for AI Chips. Asher Brass. Available at: https://www.iaps.ai/research/location- verification-for-ai-chips (Accessed: 8 August 2024). Brundage, M. et al. (2018) ‘The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mi...
2024 arXiv
-
[14]
Brundage, M. et al. (2020) ‘Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims’, arXiv:2004.07213 [cs] [Preprint]. Available at: http://arxiv.org/abs/2004.07213 (Accessed: 23 November 2021). CDEI (2021) The roadmap to an effective AI assurance ecosy...
2020 arXiv
-
[15]
(2024) ‘Towards an international regulatory framework for AI safety: lessons from the IAEA’s nuclear safety regulations’, Humanities and Social Sciences Communications, 11(1), pp
Cha, S. (2024) ‘Towards an international regulatory framework for AI safety: lessons from the IAEA’s nuclear safety regulations’, Humanities and Social Sciences Communications, 11(1), pp. 1–13. Available at: https://doi.org/10.1057/s41599-024-03017-1
2024 doi
-
[16]
(2024) Evaluating An AI Chip Registration Policy
Cheng, D. (2024) Evaluating An AI Chip Registration Policy. Convergence Analysis. Available at: https://www.convergenceanalysis.org/research/evaluating-an-ai-chip-registration-policy. 42
2024
-
[17]
(2023) ‘AI Desperately Needs Global Oversight’, Wired, 6 April
Chowdhury, R. (2023) ‘AI Desperately Needs Global Oversight’, Wired, 6 April. Available at: https://www.wired.com/story/ai-desperately-needs-global-oversight/ (Accessed: 22 April 2024)
2023
-
[18]
(2023) Trends in the dollar training cost of machine learning systems
Cottier, B. (2023) Trends in the dollar training cost of machine learning systems. Epoch. Avail- able at: https://epochai.org/blog/trends-in-the-dollar-training-cost-of-machine-learning-systems (Ac- cessed: 15 March 2023)
2023
-
[19]
and Feldgoise, J
Dohmen, H. and Feldgoise, J. (2023) ‘Explainer: The Commerce Department’s October 2023 Export Control Update – A Bigger Yard, A Higher Fence: Understanding BIS’s Expanded Controls on Advanced Computing Exports’, Center for Security and Emerging Technology, 4 December
2023
-
[20]
and Dafoe, A
Dragan, A., King, H. and Dafoe, A. (2024) Frontier Safety Framework. Google DeepMind. Available at: https://deepmind.google/discover/blog/introducing-the-frontier-safety-framework/
2024
-
[21]
and Depp, M
Drexel, B. and Depp, M. (2024) ‘Every Country Is on Its Own on AI’, Foreign Policy, 18 April. Available at: https://foreignpolicy.com/2023/06/13/ai-regulation-international-nuclear/ (Accessed: 17 April 2024). DSIT (2023) Capabilities and risks from frontier AI: A discussion pa...
2024
-
[22]
(2017) The road to prohibition: nuclear hierarchy and disarmament, 1968–2017
Egeland, K. (2017) The road to prohibition: nuclear hierarchy and disarmament, 1968–2017. http://purl.org/dc/dcmitype/Text. University of Oxford. Available at: https://ora.ox.ac.uk/objects/uuid:b03d68ab- 4748-4de7-a2e9-15616de6a05c (Accessed: 11 November 2021)
2017
-
[23]
and Lu, E
Fedasiuk, R., Elmgren, K. and Lu, E. (2022) Silicon Twist: Managing the Chinese Military’s Ac- cess to AI Chips. Center for Security and Emerging Technology. Available at: https://doi.org/10.51593/20210068
2022 doi
-
[24]
and Wenger, A
Fischer, S.-C. and Wenger, A. (2019) ‘A Neutral Hub for AI Research’. Available at: https://doi.org/10.3929/ETHZ- B-000332541
2019 doi
-
[25]
and Grunewald, E
Fist, T. and Grunewald, E. (2023) Preventing AI Chip Smuggling to China. Washington DC: CNAS. Available at: https://www.cnas.org/publications/reports/preventing-ai-chip-smuggling-to- china (Accessed: 5 December 2023)
2023
-
[26]
(2016) Who Cares about Particle Physics?: Making Sense of the Higgs Boson, the Large Hadron Collider and CERN
Gagnon, P. (2016) Who Cares about Particle Physics?: Making Sense of the Higgs Boson, the Large Hadron Collider and CERN. Oxford: OUP Oxford. Gaithersburg Md Nist (2024) Managing Misuse Risk for Dual-Use Foundation Models. NIST AI NIST AI 800-1 ipd. Gaithersburg, MD: National ...
2016 doi
-
[27]
and Aird, M
Grunewald, E. and Aird, M. (2023) AI chip smuggling into China: Potential paths, quantities, and countermeasures. Washington DC: IAPS
2023
-
[28]
and Fist, T
Grunewald, E. and Fist, T. (2024) Comments on the Advanced Computing/Supercomputing IFR: Export Control Strategy & Enforcement for AI Chips. CNAS. Available at: https://www.cnas.org/publications/commentary/comments- on-the-advanced-computing-supercomputing-ifr-export-control-s...
2024
-
[29]
Guterres, A. (2023) ‘Secretary-General Urges Security Council to Ensure Transparency, Ac- countability, Oversight, in First Debate on Artificial Intelligence — Meetings Coverage and Press Releases’, 18 July
2023
-
[30]
and Dennis, C
Hausenloy, J. and Dennis, C. (2023) Towards a UN Role in Governing Foundation Artificial Intelligence Models. United Nations University
2023
-
[31]
Hausenloy, J., Miotti, A., and Dennis, C. (2023). Multinational AGI Consortium (MAGIC): A Proposal for International Coordination on AI
2023
-
[32]
Heim, L. et al. (2024) Governing through the cloud: the intermediary role of compute providers in AI regulation. Oxford Martin AI Governance Initiative. Available at: https://www.oxfordmartin.ox.ac.uk/publications/governing- through-the-cloud-the-intermediary-role-of-compute-p...
2024
-
[33]
and Koessler, L
Heim, L. and Koessler, L. (2024) ‘Training Compute Thresholds: Features and Functions in AI Regulation’. arXiv. Available at: http://arxiv.org/abs/2405.10799 (Accessed: 8 August 2024)
2024 arXiv
- [34]
-
[35]
Hirschman, A.O. (1970). Exit, Voice, and Loyalty: Responses to Decline in Firms, Organizations, and States. Harvard University Press
1970
- [36]
-
[37]
(2024) ‘Safety cases at AISI’, AI Safety Institute, 23 August
Irving, G. (2024) ‘Safety cases at AISI’, AI Safety Institute, 23 August. Available at: https://www.aisi.gov.uk/work/safety- cases-at-aisi (Accessed: 13 September 2024)
2024
-
[38]
(2024) ‘Semiconductor CapEx Down in 2024’, Semiwiki, 3 August
Jewell, B. (2024) ‘Semiconductor CapEx Down in 2024’, Semiwiki, 3 August. Available at: https://semiwiki.com/semiconductor-services/343353-semiconductor-capex-down-in-2024/ (Accessed: 8 August 2024)
2024
-
[39]
Kemp, L. et al. (2019) UN High-level Panel on Digital Cooperation: A Proposal for Inter- national AI Governance. Available at: https://www.cser.ac.uk/news/advice-un-high-level-panel- digital-cooperation/ (Accessed: 17 April 2024)
2019
-
[40]
and Renda, A
Kerry, C.F., Meltzer, J.P. and Renda, A. (2022) Global AI Cooperation on the Ground: AI R&D on a global scale. Brookings
2022
-
[41]
and Snyder, C.M
Kremer, M., Levin, J. and Snyder, C.M. (2020) ‘Advance Market Commitments: Insights from Theory and Experience’, AEA Papers and Proceedings, 110, pp. 269–273. Available at: https://doi.org/10.1257/pandp.20201017. La Commission de l’intelligence artificielle (2024) IA: notre am...
2020 doi
-
[42]
and Ho, L
Law, H. and Ho, L. (2023) ‘Can a dual mandate be a model for the global governance of AI?’, Nature Reviews Physics, 5(12), pp. 706–707. Available at: https://doi.org/10.1038/s42254-023- 00670-4. LCFI (2023) ‘Do companies’ AI Safety Policies meet government best practice? - LCF...
2023 doi
-
[43]
Linghan, Z. et al. (2024) ‘Artificial Intelligence Law of the People’s Republic of China (Draft for Suggestions from Scholars)’, Center for Security and Emerging Technology, 2 May. Available at: https://cset.georgetown.edu/publication/china-ai-law-draft/ (Accessed: 8 August 2024)
2024
-
[44]
and Villalobos, J.J
Maas, M.M. and Villalobos, J.J. (2023) ‘International AI Institutions: A Literature Review of
2023
-
[45]
Rochester, NY
Models, Examples, and Proposals’. Rochester, NY. Available at: https://doi.org/10.2139/ssrn.4579773
-
[46]
(2023) ‘IAEA Safeguards: A Model for International AI Regulation’, iGlobe- news, 19 May
Markhof, F.M. (2023) ‘IAEA Safeguards: A Model for International AI Regulation’, iGlobe- news, 19 May. Available at: https://www.iglobenews.org/iaea-safeguards-a-model-for-international- ai-regulation/ (Accessed: 17 April 2024)
2023
-
[47]
(2023) ‘Why the IAEA model may not be best for regulating artificial intelligence’, Bulletin of the Atomic Scientists, 9 June
Mecklin, J. (2023) ‘Why the IAEA model may not be best for regulating artificial intelligence’, Bulletin of the Atomic Scientists, 9 June. Available at: https://thebulletin.org/2023/06/why-the- iaea-model-may-not-be-best-for-regulating-artificial-intelligence/ (Accessed: 17 Ap...
2023
-
[48]
Nevo, S. et al. (2024) Securing AI Model Weights: Preventing Theft and Misuse of Frontier Models. RAND Corporation. O’Keefe, C. (2024) ‘Chips for Peace: How the U.S. and Its Allies Can Lead on Safe and Beneficial AI’, Default, 10 July
2024
-
[49]
(2024) ‘Mark Zuckerberg - Llama 3, Open Sourcing $10b Models, & Caesar Augustus’, 28 March
Patel, D. (2024) ‘Mark Zuckerberg - Llama 3, Open Sourcing $10b Models, & Caesar Augustus’, 28 March. Available at: https://www.dwarkeshpatel.com/p/mark-zuckerberg (Accessed: 22 April 2024)
2024
-
[50]
(2023) ‘IAEA for AI? That won’t fly’, 18 August
Payne, K. (2023) ‘IAEA for AI? That won’t fly’, 18 August. Available at: https://www.kennethpayne.uk/p/iaea- for-ai-that-wont-fly (Accessed: 17 April 2024)
2023
-
[51]
Pouget, H. et al. (2024) The Future of International Scientific Assessments of AI’s Risks. Carnegie Endowment for International Peace. Available at: https://carnegieendowment.org/research/2024/08/the- future-of-international-scientific-assessments-of-ais-risks?lang=en
2024
-
[52]
(2023) ‘Pope Francis calls for binding global treaty to regulate AI’, Reuters, 14 Decem- ber
Pullella, P. (2023) ‘Pope Francis calls for binding global treaty to regulate AI’, Reuters, 14 Decem- ber. Available at: https://www.reuters.com/technology/pope-calls-binding-global-treaty-artificial- intelligence-2023-12-14/ (Accessed: 12 August 2024)
2023
-
[53]
(1986) The making of the atomic bomb
Rhodes, R. (1986) The making of the atomic bomb. New York: Simon & Schuster
1986
-
[54]
Roberts, H. et al. (2023) ‘Global AI governance: barriers and pathways forward’. Rochester, NY. Available at: https://doi.org/10.2139/ssrn.4588040
2023 doi
-
[55]
(2019) ‘The CERN Community; A Mechanism for Effective Global Collaboration?’, Global Policy, 10(1), pp
Robinson, M. (2019) ‘The CERN Community; A Mechanism for Effective Global Collaboration?’, Global Policy, 10(1), pp. 41–51. Available at: https://doi.org/10.1111/1758-5899.12608
2019
-
[56]
(2021) ‘Big Science Collaborations; Lessons for Global Governance and Leadership’, Global Policy, 12(1), pp
Robinson, M. (2021) ‘Big Science Collaborations; Lessons for Global Governance and Leadership’, Global Policy, 12(1), pp. 66–80
2021
-
[57]
(2023) ‘Artificial intelligence and biological misuse: Differentiating risks of lan- guage models and biological design tools’
Sandbrink, J.B. (2023) ‘Artificial intelligence and biological misuse: Differentiating risks of lan- guage models and biological design tools’. arXiv. Available at: https://doi.org/10.48550/arXiv.2306.13952
- [58]
-
[59]
Sevilla, J. et al. (2024) Can AI Scaling Continue Through 2030? Epoch AI. Available at: https://epochai.org/blog/can-ai-scaling-continue-through-2030 (Accessed: 21 August 2024)
2024
-
[60]
(2023) ‘What does it take to catch a Chinchilla? Verifying Rules on Large-Scale Neural Network Training via Compute Monitoring’
Shavit, Y. (2023) ‘What does it take to catch a Chinchilla? Verifying Rules on Large-Scale Neural Network Training via Compute Monitoring’. arXiv. Available at: https://doi.org/10.48550/arXiv.2303.11341
-
[61]
(2023) China’s AI Regulations and How They Get Made
Sheehan, M. (2023) China’s AI Regulations and How They Get Made. Washington DC: Carnegie Endowment for International Peace
2023
- [62]
-
[63]
and Trager, R.F
Stafford, E. and Trager, R.F. (2022) The IAEA Solution: Knowledge Sharing to Prevent Danger- ous Technology Races. Centre for the Governance of AI (GovAI). Available at: https://www.governance.ai/research- paper/knowledge-sharing-to-prevent-dangerous-technology-races
2022
-
[64]
(2022) ‘Foundations for the future: institution building for the purpose of artificial intelli- gence governance’, AI and Ethics, 2(3), pp
Stix, C. (2022) ‘Foundations for the future: institution building for the purpose of artificial intelli- gence governance’, AI and Ethics, 2(3), pp. 463–476. Available at: https://doi.org/10.1007/s43681- 021-00093-w. The Elders (2023) ‘The Elders urge global co-operation to ma...
2022 doi
-
[65]
Trager, R. et al. (2023) ‘International Governance of Civilian AI: A Jurisdictional Certification Approach’. arXiv. United States Department of Justice Office of Public Affairs (2023) Justice and Commerce De- partments Announce Creation of Disruptive Technology Strike Force. U...
2023
-
[66]
(2023) ‘Opinion — IAEA for AI? That Model Has Already Failed’, Wall Street Journal, 1 June
Watson, M. (2023) ‘Opinion — IAEA for AI? That Model Has Already Failed’, Wall Street Journal, 1 June. Available at: https://www.wsj.com/articles/iaea-for-ai-that-model-has-already- failed-chaptgpt-technology-nuclear-proliferation-4339543b (Accessed: 17 April 2024)
2023
-
[67]
CERN for AI
Wold, J.W. (2024) ‘Von der Leyen gives nod to€100 billion “CERN for AI” proposal’, www.euractiv.com, 25 July. Available at: https://www.euractiv.com/ section/digital/news/von-der-leyen-gives-nod-to- e100-billion-cern-for-ai-proposal/ (Accessed: 8 August 2024)
2024
-
[68]
Zhang, D. et al. (2022) Enhancing International Cooperation in AI Research: The Case for a Mul- tilateral AI Research Institute. Stanford Institute for Human-Centered Artificial Intelligence. Avail- able at: https://hai.stanford.edu/white-paper-enhancing-international-cooperat...
2022
Reviewed August 6, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.