Visual summary of operating lessons from Miles Brundage.

Lessons from Miles Brundage

Miles Brundage announced on Miles’s Substack the co-founding of a non-profit focused on frontier AI auditing. The items below are things Brundage wrote in Substack posts about AI policy, security, auditing and governance. — “AI Auditing 2027,” Miles Brundage (Miles’s Substack, July 16, 2025).

Part 1: AGI Readiness and Safety

  1. On a Major Discovery: In the Substack post “Time's Up for AI Policy” (December 20, 2024), Brundage wrote: "The fact that this all works so well — and so much more easily and quickly than many expected — is easily one of the biggest and most important discoveries in human history, and still not fully appreciated". — “Time's Up for AI Policy,” Miles Brundage (Miles’s Substack, December 20, 2024).
  2. On Not Having an Informed Opinion: In the Substack post “Should AI Progress Speed Up, Slow Down, or Stay the Same?” (November 2, 2024), Brundage wrote: "you don’t have (and no one has) a super informed opinion on the answer to the question in the title". — “Should AI Progress Speed Up, Slow Down, or Stay the Same?,” Miles Brundage (Miles’s Substack, November 2, 2024)
  3. On Remembering How Wild It Is: In the Substack post “Time's Up for AI Policy” (December 20, 2024), Brundage wrote: "Refusing to forget how wild it is that AI capabilities are what they are". — “Time's Up for AI Policy,” Miles Brundage (Miles’s Substack, December 20, 2024).
  4. On Benchmarks Saturating: In the Substack post “Should AI Progress Speed Up, Slow Down, or Stay the Same?” (November 2, 2024), Brundage wrote: "AI systems are starting to outperform experts on test questions in areas like physics, chemistry, and coding". — “Should AI Progress Speed Up, Slow Down, or Stay the Same?,” Miles Brundage (Miles’s Substack, November 2, 2024)
  5. On No Human-Level Ceiling: In the Substack post “Time's Up for AI Policy” (December 20, 2024), Brundage wrote: "Recognizing that there is much further to go, and no obvious "human-level" ceiling". — “Time's Up for AI Policy,” Miles Brundage (Miles’s Substack, December 20, 2024).
  6. On Hard Questions Integrating: In the Substack post “Should AI Progress Speed Up, Slow Down, or Stay the Same?” (November 2, 2024), Brundage wrote: "These individual questions are hard on their own, and integrating them into an overall framework to conclude something useful about the ideal pace of AI progress is also hard". — “Should AI Progress Speed Up, Slow Down, or Stay the Same?,” Miles Brundage (Miles’s Substack, November 2, 2024)
  7. On a Moral Obligation: In the Substack post “Time's Up for AI Policy” (December 20, 2024), Brundage wrote: "Taking seriously one's moral obligation to shape the outcomes of AGI as positively as one can". — “Time's Up for AI Policy,” Miles Brundage (Miles’s Substack, December 20, 2024).
  8. On Brakes: In the Substack post “Should AI Progress Speed Up, Slow Down, or Stay the Same?” (November 2, 2024), Brundage wrote: "By brakes, I mean well-specified and well-analyzed technical and policy options for slowing AI progress if it turns out that – based on answers to questions like those above – it makes sense to do so". — “Should AI Progress Speed Up, Slow Down, or Stay the Same?,” Miles Brundage (Miles’s Substack, November 2, 2024)
  9. On Tricky Progress: In the Substack post “We're in Triage Mode for AI Policy” (February 22, 2026), Brundage wrote: "It will already be tricky to get significant, rapid progress on the basics of AI safety and security and avoiding extreme concentration of power due to AI". — “We're in Triage Mode for AI Policy,” Miles Brundage (Miles’s Substack, February 22, 2026).
  10. On Preparing for Scenarios: In the Substack post “Should AI Progress Speed Up, Slow Down, or Stay the Same?” (November 2, 2024), Brundage wrote: "Hopefully policymakers get on top of things, we soon see evidence that safety is easy, etc., but we don’t know that those things will happen, so let’s be prepared for various scenarios". — “Should AI Progress Speed Up, Slow Down, or Stay the Same?,” Miles Brundage (Miles’s Substack, November 2, 2024)
  11. On Triage: In the Substack post “We're in Triage Mode for AI Policy” (February 22, 2026), Brundage wrote: "Unfortunately, triaging is needed". — “We're in Triage Mode for AI Policy,” Miles Brundage (Miles’s Substack, February 22, 2026).
  12. On Simple Solutions to Hard Problems: In the Substack post “Should AI Progress Speed Up, Slow Down, or Stay the Same?” (November 2, 2024), Brundage wrote: "People see the problems, and look around for a solution, and come up with something simple, but policymaking is hard and requires foresight, debate, and serious research". — “Should AI Progress Speed Up, Slow Down, or Stay the Same?,” Miles Brundage (Miles’s Substack, November 2, 2024)
  13. On Stumbling Into Avoidable Harm: In the Substack post “We're in Triage Mode for AI Policy” (February 22, 2026), Brundage wrote: "But if we don’t do much more than we’re currently doing, we’ll almost certainly stumble into a lot of bad things that we could have avoided, and possibly hit one of the worst case scenarios". — “We're in Triage Mode for AI Policy,” Miles Brundage (Miles’s Substack, February 22, 2026).
  14. On Paying for Thinking: In the Substack post “AI is a Liquid, Not a Solid” (June 7, 2025), Brundage wrote: "The amount of thinking that you get out of AI increasingly depends on how much you pay". — “AI is a Liquid, Not a Solid,” Miles Brundage (Miles’s Substack, June 7, 2025).
  15. On Running Models: In the Substack post “AI Policy Considerations for the Trump Administration: Part I - Navigating Innovation and Safety” (November 12, 2024), Brundage wrote: "It is becoming increasingly clear that leadership in AI is not just a matter of training big models but also investing a lot of computing power into running them". — “AI Policy Considerations for the Trump Administration: Part I - Navigating Innovation and Safety,” Miles Brundage (Miles’s Substack, November 12, 2024)
  16. On Liquid Access: In the Substack post “AI is a Liquid, Not a Solid” (June 7, 2025), Brundage wrote: "In addition to AI quantities being liquid-like, AI access is also liquid-like". — “AI is a Liquid, Not a Solid,” Miles Brundage (Miles’s Substack, June 7, 2025).
  17. On the AI Brain of Government: In the Substack post “AI Policy Considerations for the Trump Administration: Part I - Navigating Innovation and Safety” (November 12, 2024), Brundage wrote: "Currently, the US government’s “AI brain” is the US AI Safety Institute in the Department of Commerce". — “AI Policy Considerations for the Trump Administration: Part I - Navigating Innovation and Safety,” Miles Brundage (Miles’s Substack, November 12, 2024)
  18. On Ease of Spreading: In the Substack post “AI is a Liquid, Not a Solid” (June 7, 2025), Brundage wrote: "our policies and our overall mental model of the technology needs to reflect AI’s ease of spreading". — “AI is a Liquid, Not a Solid,” Miles Brundage (Miles’s Substack, June 7, 2025).

Part 2: The Malicious Use of AI

  1. On Being Behind on Security: In the Substack post “Why Security Comes First” (January 10, 2025), Brundage wrote: "We are arguably already way behind in terms of security relative to the importance of AI as a technology". — “Why Security Comes First,” Miles Brundage (Miles’s Substack, January 10, 2025).
  2. On Cooperative Compromise: In the Substack post “Reasons that AI Can Help with AI Governance” (November 14, 2024), Brundage wrote: "These could reduce transaction costs (for finding a cooperative solution) and provide a workaround to the issue of asymmetrical information, both of which can be major drivers of conflicts between people and groups". — “Reasons that AI Can Help with AI Governance,” Miles Brundage (Miles’s Substack, November 14, 2024)
  3. On Governing a Technology That Moves: In the Substack post “Why Security Comes First” (January 10, 2025), Brundage wrote: "You can’t govern a technology if it won’t sit in one place, and you can’t make that technology controllable if random bad actors are tampering with it, poisoning its data, etc". — “Why Security Comes First,” Miles Brundage (Miles’s Substack, January 10, 2025).
  4. On Advice During Disruption: In the Substack post “Reasons that AI Can Help with AI Governance” (November 14, 2024), Brundage wrote: "While everything is getting disrupted (in good and bad ways) by AI, human decision-makers will have an increasingly intelligent source of advice during this process – AI". — “Reasons that AI Can Help with AI Governance,” Miles Brundage (Miles’s Substack, November 14, 2024)
  5. On AI Years: In the Substack post “Why Security Comes First” (January 10, 2025), Brundage wrote: "One calendar year is a long time in “AI years,” so the current situation is just unreasonable at a societal level". — “Why Security Comes First,” Miles Brundage (Miles’s Substack, January 10, 2025).
  6. On Open Source Benefits: In the Substack post “AI Policy Considerations for the Trump Administration: Part I - Navigating Innovation and Safety” (November 12, 2024), Brundage wrote: "Open source AI has many benefits, fostering innovation and potentially advancing the US’s competitive position". — “AI Policy Considerations for the Trump Administration: Part I - Navigating Innovation and Safety,” Miles Brundage (Miles’s Substack, November 12, 2024)
  7. On Access Not Always Being Good: In the Substack post “Why Security Comes First” (January 10, 2025), Brundage wrote: "Sometimes it’s good to give everyone access to a technology. But that will not always be a good idea". — “Why Security Comes First,” Miles Brundage (Miles’s Substack, January 10, 2025).
  8. On Human-Looking Outputs: In the Substack post “Scoring Humanity’s Progress on AI Governance” (October 23, 2024), Brundage wrote: "We used to have a world where we could be totally sure human-looking outputs were human-generated". — “Scoring Humanity’s Progress on AI Governance,” Miles Brundage (Miles’s Substack, October 23, 2024)
  9. On Rigor Sometimes: In the Substack post “Why I'm Not a Security Doomer” (March 11, 2025), Brundage wrote: "The question is not whether we should apply very rigorous standards of security to all systems but whether we should do it sometimes – and to figure out how to do it at the lowest possible cost". — “Why I'm Not a Security Doomer,” Miles Brundage (Miles’s Substack, March 11, 2025).
  10. On Security Failures: In the Substack post “Why I'm Not a Security Doomer” (March 11, 2025), Brundage wrote: "There are countless examples of information security failures – because it’s a hard problem". — “Why I'm Not a Security Doomer,” Miles Brundage (Miles’s Substack, March 11, 2025).
  11. On Hard but Not Impossible: In the Substack post “Why I'm Not a Security Doomer” (March 11, 2025), Brundage wrote: "This is not by any means trivial but I have yet to see an argument as to why it’s impossible". — “Why I'm Not a Security Doomer,” Miles Brundage (Miles’s Substack, March 11, 2025).
  12. On Too Little Butter: In the Substack post “The Bread and Butter Problem in AI Policy” (November 6, 2024), Brundage wrote: "there is too little safety and security “butter” spread over too much AI development/deployment “bread". — “The Bread and Butter Problem in AI Policy,” Miles Brundage (Miles’s Substack, November 6, 2024).
  13. On a Tiny Fraction of Talent: In the Substack post “The Bread and Butter Problem in AI Policy” (November 6, 2024), Brundage wrote: "only a tiny fraction of the industry’s talent is applied towards making it safe and secure, and only a tiny amount of time is spent doing so, compared to what would arguably be societally optimal considering the risks in question". — “The Bread and Butter Problem in AI Policy,” Miles Brundage (Miles’s Substack, November 6, 2024).

Part 3: Compute and Governance

  1. On Third-Party Verification: In the Substack post “The Launch of AVERI” (January 15, 2026), Brundage wrote: "rigorous third-party verification of frontier AI developers’ safety and security claims, and evaluation of their systems and practices against relevant standards, based on deep, secure access to non-public information". — “The Launch of AVERI,” Miles Brundage (Miles’s Substack, January 15, 2026).
  2. On Limited Access for Auditors: In the Substack post “AI Auditing 2027” (July 16, 2025), Brundage wrote: "Today, third parties sometimes try to do something like auditing, but they have limited access and resources, and AI companies have limited incentives to engage with them". — “AI Auditing 2027,” Miles Brundage (Miles’s Substack, July 16, 2025).
  3. On Black-Box Product Testing: In the Substack post “AI Auditing 2027” (July 16, 2025), Brundage wrote: "The state-of-the-art today can be more accurately described as “last-minute, black-box product testing” rather than auditing". — “AI Auditing 2027,” Miles Brundage (Miles’s Substack, July 16, 2025).
  4. On What Auditing Is: In the Substack post “AI Auditing 2027” (July 16, 2025), Brundage wrote: "rigorous, independent, and well-resourced assessment of AI systems and organizations against a set of safety and security standards". — “AI Auditing 2027,” Miles Brundage (Miles’s Substack, July 16, 2025).
  5. On Public Safety and Adoption: In the Substack post “AI Auditing 2027” (July 16, 2025), Brundage wrote: "this situation threatens public safety, international stability, and sustainable commercial adoption of AI". — “AI Auditing 2027,” Miles Brundage (Miles’s Substack, July 16, 2025).
  6. On Making Auditing Universal: In the Substack post “AI Auditing 2027” (July 16, 2025), Brundage wrote: "I’m co-founding a non-profit focused on making frontier AI auditing effective and universal". — “AI Auditing 2027,” Miles Brundage (Miles’s Substack, July 16, 2025).
  7. On Individual Systems: In the Substack post “Why We Need to Think Bigger in AI Policy (Literally)” (May 18, 2025), Brundage wrote: "Most (though not all) thinking on AI safety, security, and policy is framed in terms of mitigating the risks of individual AI systems". — “Why We Need to Think Bigger in AI Policy (Literally),” Miles Brundage (Miles’s Substack, May 18, 2025).
  8. On How a Model Is Used: In the Substack post “Why We Need to Think Bigger in AI Policy (Literally)” (May 18, 2025), Brundage wrote: "How a model is used as part of a larger system, including how much test-time compute is used, matters a lot". — “Why We Need to Think Bigger in AI Policy (Literally),” Miles Brundage (Miles’s Substack, May 18, 2025).
  9. On Perfect and Good: In the Substack post “Why We Need to Think Bigger in AI Policy (Literally)” (May 18, 2025), Brundage wrote: "we shouldn’t let perfect be the enemy of good. But we need to make sure we’re heading in the right general direction". — “Why We Need to Think Bigger in AI Policy (Literally),” Miles Brundage (Miles’s Substack, May 18, 2025).
  10. On Safety Talent Stretched Thin: In the Substack post “The Bread and Butter Problem in AI Policy” (November 6, 2024), Brundage wrote: "At a global level, the people working to ensure that AI systems are safe and don’t get stolen are stretched thin, and new companies popping up all the time make this worse". — “The Bread and Butter Problem in AI Policy,” Miles Brundage (Miles’s Substack, November 6, 2024).

Part 4: Verifiable Claims and Trustworthy AI

  1. On Verifying Claims With Ombudspeople: In the Substack post “The Human Bridge to Trustworthy AI: Independent Ombudspeople Could Help Verify Claims by Companies and Countries about AI” (January 7, 2025), Brundage wrote: "An organization could be created in order to verify the claims made by companies and countries about AI – and that’s it". — “The Human Bridge to Trustworthy AI: Independent Ombudspeople Could Help Verify Claims by Companies and Countries about AI,” Miles Brundage (Miles’s Substack, January 7, 2025).
  2. On Voluntary Services: In the Substack post “The Human Bridge to Trustworthy AI: Independent Ombudspeople Could Help Verify Claims by Companies and Countries about AI” (January 7, 2025), Brundage wrote: "Using GAAIO’s services would be totally voluntary, its conclusions would have no force beyond what governments choose to do with them". — “The Human Bridge to Trustworthy AI: Independent Ombudspeople Could Help Verify Claims by Companies and Countries about AI,” Miles Brundage (Miles’s Substack, January 7, 2025).
  3. On Inspectors’ Body and Brain: In the Substack post “The Human Bridge to Trustworthy AI: Independent Ombudspeople Could Help Verify Claims by Companies and Countries about AI” (January 7, 2025), Brundage wrote: "While IAEA inspectors have fancy tools, their most important “tools” are their body and brain". — “The Human Bridge to Trustworthy AI: Independent Ombudspeople Could Help Verify Claims by Companies and Countries about AI,” Miles Brundage (Miles’s Substack, January 7, 2025).
  4. On a Sketch to Spark Thinking: In the Substack post “The Human Bridge to Trustworthy AI: Independent Ombudspeople Could Help Verify Claims by Companies and Countries about AI” (January 7, 2025), Brundage wrote: "I’m not sure about the exact details of this idea being right, but hopefully the sketch of one solution below will spark some creative thinking on the overall problem". — “The Human Bridge to Trustworthy AI: Independent Ombudspeople Could Help Verify Claims by Companies and Countries about AI,” Miles Brundage (Miles’s Substack, January 7, 2025).
  5. On Fines for Violators: In the Substack post “Standards, Incentives, and Evidence: The Frontier AI Governance Triad” (June 19, 2025), Brundage wrote: "My personal preferred path for incentivizing compliance with AI standards is for the US federal government to pass legislation that creates fines for violators of a specific set of safety and security standards". — “Standards, Incentives, and Evidence: The Frontier AI Governance Triad,” Miles Brundage (Miles’s Substack, June 19, 2025).
  6. On Market-Based Incentives: In the Substack post “Standards, Incentives, and Evidence: The Frontier AI Governance Triad” (June 19, 2025), Brundage wrote: "we should not just focus on governments but also explore a range of market-based incentives, as well as incentives tied to specific high-risk industries". — “Standards, Incentives, and Evidence: The Frontier AI Governance Triad,” Miles Brundage (Miles’s Substack, June 19, 2025).
  7. On a Governance Triad: In the Substack post “Standards, Incentives, and Evidence: The Frontier AI Governance Triad” (June 19, 2025), Brundage wrote: "This triad is not exhaustive". — “Standards, Incentives, and Evidence: The Frontier AI Governance Triad,” Miles Brundage (Miles’s Substack, June 19, 2025).
  8. On Policymakers Stretched Thin: In the Substack post “The Bread and Butter Problem in AI Policy” (November 6, 2024), Brundage wrote: "Policymakers will also soon be stretched thin as they assess risks related to, and implement reporting requirements on, zillions of companies, models, and products". — “The Bread and Butter Problem in AI Policy,” Miles Brundage (Miles’s Substack, November 6, 2024).
  9. On Smuggled Assumptions: In the Substack post “The Bread and Butter Problem in AI Policy” (November 6, 2024), Brundage wrote: "I don’t think that this conclusion obviously follows from the bread and butter metaphor — there are many assumptions being smuggled in". — “The Bread and Butter Problem in AI Policy,” Miles Brundage (Miles’s Substack, November 6, 2024).

Part 5: Expert Red Teaming and Evaluation

  1. On What a Megapaper Is: In the Substack post “Megapapers and How to Write Them” (August 27, 2025), Brundage wrote: "By megapaper, I mean a research paper/report that involves a large number of authors (roughly 20+), institutions (roughly 15+), and disciplines (roughly 10+)". — “Megapapers and How to Write Them,” Miles Brundage (Miles’s Substack, August 27, 2025).
  2. On Diverse Authors: In the Substack post “Megapapers and How to Write Them” (August 27, 2025), Brundage wrote: "the number and diversity of authors is, critically, not just for show but a reflection of the range of expertise brought to bear on the paper’s overall thesis and various specific claims". — “Megapapers and How to Write Them,” Miles Brundage (Miles’s Substack, August 27, 2025).
  3. On Intellectual Megaprojects: In the Substack post “Megapapers and How to Write Them” (August 27, 2025), Brundage wrote: "Megapapers are a specific type of “intellectual megaproject,”". — “Megapapers and How to Write Them,” Miles Brundage (Miles’s Substack, August 27, 2025).
  4. On Wikipedia as a Megaproject: In the Substack post “Megapapers and How to Write Them” (August 27, 2025), Brundage wrote: "Wikipedia is a much larger intellectual megaproject than any megapaper". — “Megapapers and How to Write Them,” Miles Brundage (Miles’s Substack, August 27, 2025).
  5. On Bare Minimum Standards: In the Substack post “Iran and the opportunities and limits of an “IAEA for AI”” (June 18, 2025), Brundage wrote: "So it may be possible to negotiate some set of bare minimum safety and security standards, and then audit everyone (at least everyone who has very large amounts of computing power) against those standards". — “Iran and the opportunities and limits of an “IAEA for AI”,” Miles Brundage (Miles’s Substack, June 18, 2025).
  6. On Breathing Room: In the Substack post “Iran and the opportunities and limits of an “IAEA for AI”” (June 18, 2025), Brundage wrote: "Breathing room in an otherwise very competitive situation is valuable". — “Iran and the opportunities and limits of an “IAEA for AI”,” Miles Brundage (Miles’s Substack, June 18, 2025).
  7. On a World That Agrees: In the Substack post “Iran and the opportunities and limits of an “IAEA for AI”” (June 18, 2025), Brundage wrote: "One reason that an IAEA for AI might be necessary eventually is that the world actually agrees on a lot of things related to AI". — “Iran and the opportunities and limits of an “IAEA for AI”,” Miles Brundage (Miles’s Substack, June 18, 2025).
  8. On Credibly Signaling Benignness: In the Substack post “Finding Ways to Credibly Signal the Benignness of AI Development and Deployment is an Urgent Priority” (October 28, 2024), Brundage wrote: "It’s hard to credibly signal the benignness of AI development and deployment, but not obviously impossible". — “Finding Ways to Credibly Signal the Benignness of AI Development and Deployment is an Urgent Priority,” Miles Brundage (Miles’s Substack, October 28, 2024).
  9. On Militaries Signaling Benignness: In the Substack post “Finding Ways to Credibly Signal the Benignness of AI Development and Deployment is an Urgent Priority” (October 28, 2024), Brundage wrote: "Internationally, militaries should seek to credibly signal benignness and should demand the same of their adversaries". — “Finding Ways to Credibly Signal the Benignness of AI Development and Deployment is an Urgent Priority,” Miles Brundage (Miles’s Substack, October 28, 2024).
  10. On Chemical Inspections: In the Substack post “Finding Ways to Credibly Signal the Benignness of AI Development and Deployment is an Urgent Priority” (October 28, 2024), Brundage wrote: "inspections of chemical factories routinely take place in order to ensure that chemical weapons are not being produced, and this is done in a way that protects other important interests like intellectual property and innovation". — “Finding Ways to Credibly Signal the Benignness of AI Development and Deployment is an Urgent Priority,” Miles Brundage (Miles’s Substack, October 28, 2024).

Part 6: Policy, Regulation, and Institutions

  1. On Incentives and Corner-Cutting: In the Substack post “Some Very Important Things (That I Won't Be Working On This Year)” (February 15, 2025), Brundage wrote: "Contrary to some people’s beliefs, companies do not in fact have sufficient incentives to mitigate all major risks, and competition is driving corner-cutting that needs to be reined in somehow". — “Some Very Important Things (That I Won't Be Working On This Year),” Miles Brundage (Miles’s Substack, February 15, 2025).
  2. On Job Losses: In the Substack post “Some Very Important Things (That I Won't Be Working On This Year)” (February 15, 2025), Brundage wrote: "I’m quite confident that AI will eliminate many millions of people’s jobs in the next few years". — “Some Very Important Things (That I Won't Be Working On This Year),” Miles Brundage (Miles’s Substack, February 15, 2025).
  3. On Stop Lights for Agents: In the Substack post “Some Very Important Things (That I Won't Be Working On This Year)” (February 15, 2025), Brundage wrote: "What are the AI agent equivalents of stop lights, railroad tracks, etc. – infrastructure that we need in order to keep a powerful new technology “on the rails” while reaping its benefits?". — “Some Very Important Things (That I Won't Be Working On This Year),” Miles Brundage (Miles’s Substack, February 15, 2025).
  4. On Militaries Testing Each Other: In the Substack post “Finding Ways to Credibly Signal the Benignness of AI Development and Deployment is an Urgent Priority” (October 28, 2024), Brundage wrote: "We may eventually get to a point where militaries directly test one another’s AI systems, given that analogous things have happened in a nuclear context". — “Finding Ways to Credibly Signal the Benignness of AI Development and Deployment is an Urgent Priority,” Miles Brundage (Miles’s Substack, October 28, 2024).
  5. On Not a Complete Solution: In the Substack post “Finding Ways to Credibly Signal the Benignness of AI Development and Deployment is an Urgent Priority” (October 28, 2024), Brundage wrote: "To be clear, I don’t think that credibly signaling the benignness of AI development and deployment is a complete solution for AI policy". — “Finding Ways to Credibly Signal the Benignness of AI Development and Deployment is an Urgent Priority,” Miles Brundage (Miles’s Substack, October 28, 2024).
  6. On Rule of Law Jurisdictions: In the Substack post “Finding Ways to Credibly Signal the Benignness of AI Development and Deployment is an Urgent Priority” (October 28, 2024), Brundage wrote: "It may be advantageous from the perspective of signaling benignness if the most dangerous AI activities were to happen in jurisdictions with strong rule of law and strong transparency". — “Finding Ways to Credibly Signal the Benignness of AI Development and Deployment is an Urgent Priority,” Miles Brundage (Miles’s Substack, October 28, 2024).
  7. On AI Helping Govern AI: In the Substack post “Reasons that AI Can Help with AI Governance” (November 14, 2024), Brundage wrote: "Insofar as there are significant ways that AI can be usefully deployed in service of governing AI, the overall AI governance challenge will be easier than it otherwise would be, and I think there are many such ways". — “Reasons that AI Can Help with AI Governance,” Miles Brundage (Miles’s Substack, November 14, 2024).
  8. On Monitoring for Misuse: In the Substack post “Reasons that AI Can Help with AI Governance” (November 14, 2024), Brundage wrote: "It can help monitor for misuse and other safety issues associated with AI as it’s being used". — “Reasons that AI Can Help with AI Governance,” Miles Brundage (Miles’s Substack, November 14, 2024).
  9. On Good Guys With AI: In the Substack post “Reasons that AI Can Help with AI Governance” (November 14, 2024), Brundage wrote: "This sounds nice but embeds a bunch of assumptions about offense-defense balance scaling, the resources available to different actors, etc". — “Reasons that AI Can Help with AI Governance,” Miles Brundage (Miles’s Substack, November 14, 2024).

Part 7: International Cooperation and Geopolitics

  1. On Bombs Still Falling: In the Substack post “Iran and the opportunities and limits of an “IAEA for AI”” (June 18, 2025), Brundage wrote: "Despite the actual IAEA arguably having these properties to a large extent, and no one really doubting their top-line conclusions about Iran, bombs are still falling". — “Iran and the opportunities and limits of an “IAEA for AI”,” Miles Brundage (Miles’s Substack, June 18, 2025).
  2. On Monitoring Computing: In the Substack post “Iran and the opportunities and limits of an “IAEA for AI”” (June 18, 2025), Brundage wrote: "I think computing supply chains are much easier to monitor than nuclear ones". — “Iran and the opportunities and limits of an “IAEA for AI”,” Miles Brundage (Miles’s Substack, June 18, 2025).
  3. On Not Wanting AI to Go Rogue: In the Substack post “Iran and the opportunities and limits of an “IAEA for AI”” (June 18, 2025), Brundage wrote: "People don’t generally want AI to go rogue". — “Iran and the opportunities and limits of an “IAEA for AI”,” Miles Brundage (Miles’s Substack, June 18, 2025).
  4. On Nobody Being Ready: In the Substack post “Why I’m Leaving OpenAI and What I’m Doing Next” (October 23, 2024), Brundage wrote: "In short, neither OpenAI nor any other frontier lab is ready, and the world is also not ready". — “Why I’m Leaving OpenAI and What I’m Doing Next,” Miles Brundage (Miles’s Substack, October 23, 2024).
  5. On Railroads and Electricity: In the Substack post “Why I’m Leaving OpenAI and What I’m Doing Next” (October 23, 2024), Brundage wrote: "As with railroads, electricity, etc., corporate and government policies will be critical to ensuring safe and fair outcomes". — “Why I’m Leaving OpenAI and What I’m Doing Next,” Miles Brundage (Miles’s Substack, October 23, 2024).
  6. On Opportunity Costs: In the Substack post “Why I’m Leaving OpenAI and What I’m Doing Next” (October 23, 2024), Brundage wrote: "The opportunity costs have become very high". — “Why I’m Leaving OpenAI and What I’m Doing Next,” Miles Brundage (Miles’s Substack, October 23, 2024).
  7. On Demonstrating Compliance: In the Substack post “Why I’m Leaving OpenAI and What I’m Doing Next” (October 23, 2024), Brundage wrote: "The difficulty of demonstrating compliance without compromising sensitive information is a major barrier to arms control agreements, which requires innovation to address". — “Why I’m Leaving OpenAI and What I’m Doing Next,” Miles Brundage (Miles’s Substack, October 23, 2024).
  8. On Urgent Frontier Safety and Security: In the Substack post “Why I’m Leaving OpenAI and What I’m Doing Next” (October 23, 2024), Brundage wrote: "improving frontier AI safety and security is quite urgent, given the number of companies (dozens) that will soon (next few years at most) have systems capable of posing catastrophic risks". — “Why I’m Leaving OpenAI and What I’m Doing Next,” Miles Brundage (Miles’s Substack, October 23, 2024).
  9. On Compute Actually Used for This: In the Substack post “Reasons that AI Can Help with AI Governance” (November 14, 2024), Brundage wrote: "you need to make sure that a sizable fraction of the world’s compute is actually being used for this purpose, and not just helping with homework and ads and generating dumb blog post graphics, etc". — “Reasons that AI Can Help with AI Governance,” Miles Brundage (Miles’s Substack, November 14, 2024).

Part 8: The Role of Independent Research

  1. On Employees Getting Involved: In the Substack post “My speech at Borgo Laudato Si’” (July 16, 2026), Brundage wrote: "My emphasis on the need for AI company employees to get involved is no less urgent today than it was a few days ago". — “My speech at Borgo Laudato Si’,” Miles Brundage (Miles’s Substack, July 16, 2026).
  2. On Seven Minutes: In the Substack post “My speech at Borgo Laudato Si’” (July 16, 2026), Brundage wrote: "I only had 7 minutes so had to be quite aggressive in editing". — “My speech at Borgo Laudato Si’,” Miles Brundage (Miles’s Substack, July 16, 2026).
  3. On Not Being Neutral in Industry: In the Substack post “FAQs and General Advice on AI Policy Careers” (October 23, 2024), Brundage wrote: "If you’re in industry, it is hard for you to be seen as neutral — maybe informed, but not neutral — whereas this is possible to a greater extent in academia and civil society". — “FAQs and General Advice on AI Policy Careers,” Miles Brundage (Miles’s Substack, October 23, 2024).
  4. On Too Early and Too Late: In the Substack post “FAQs and General Advice on AI Policy Careers” (October 23, 2024), Brundage wrote: "Work can be too early, in that it’s too abstract to be useful for decision-making by any particular actor; and it can be too late, in that it’s specific but it missed the window of opportunity during which it would have been the most helpful". — “FAQs and General Advice on AI Policy Careers,” Miles Brundage (Miles’s Substack, October 23, 2024).
  5. On Distant Ideas: In the Substack post “FAQs and General Advice on AI Policy Careers” (October 23, 2024), Brundage wrote: "The most impactful ideas often come from connections and analogies between very “distant” ideas/literatures/disciplines". — “FAQs and General Advice on AI Policy Careers,” Miles Brundage (Miles’s Substack, October 23, 2024).
  6. On Not Over-Anchoring: In the Substack post “FAQs and General Advice on AI Policy Careers” (October 23, 2024), Brundage wrote: "Don’t over-anchor on any particular piece of feedback, either, and remember that no one’s perfect/everyone had to start somewhere". — “FAQs and General Advice on AI Policy Careers,” Miles Brundage (Miles’s Substack, October 23, 2024).
  7. On Governance Being Doable: In the Substack post “Scoring Humanity’s Progress on AI Governance” (October 23, 2024), Brundage wrote: "Succeeding at AI governance (not merely avoiding catastrophe but also building a much better and fairer world) seems doable, and is not especially mysterious–it just takes a lot of work". — “Scoring Humanity’s Progress on AI Governance,” Miles Brundage (Miles’s Substack, October 23, 2024).
  8. On Shared Beliefs and Cooperation: In the Substack post “Scoring Humanity’s Progress on AI Governance” (October 23, 2024), Brundage wrote: "shared beliefs in high collective upsides and high collective downsides from AI make cooperation more likely". — “Scoring Humanity’s Progress on AI Governance,” Miles Brundage (Miles’s Substack, October 23, 2024).
  9. On Fortifying Society: In the Substack post “Scoring Humanity’s Progress on AI Governance” (October 23, 2024), Brundage wrote: "If we don’t leverage our advantages and fortify society, we will eventually be cognitively outmatched by the first unintentionally or intentionally misaligned AGIs that arise, and/or they will be used for really bad purposes". — “Scoring Humanity’s Progress on AI Governance,” Miles Brundage (Miles’s Substack, October 23, 2024).