Michelle Nie Duncan (@michellesnie) header
Michelle Nie Duncan (@michellesnie) profile picture

AI, compute governance, and national security @CNASdc | Co-founder of WISE | personal views only

Washington, DC
Joined May 2014
Photo by Michelle Nie Duncan (@michellesnie)Photo by Michelle Nie Duncan (@michellesnie)Photo by Michelle Nie Duncan (@michellesnie)Photo by Michelle Nie Duncan (@michellesnie)Photo by Michelle Nie Duncan (@michellesnie)Photo by Michelle Nie Duncan (@michellesnie)Photo by Michelle Nie Duncan (@michellesnie)Photo by Michelle Nie Duncan (@michellesnie)Photo by Michelle Nie Duncan (@michellesnie)Photo by Michelle Nie Duncan (@michellesnie)
Michelle Nie Duncan retweeted
Want to work on AI policy in Washington? We're hiring! The CNAS AI Policy Lab has research and business positions open at multiple levels. We're looking for extraordinary individuals who are passionate about AI policy. Check out job opportunities at cnas.org/careers
5
11
49
3,102
Michelle Nie Duncan retweeted
i talk to a lot of people who are trying to break into ai safety, maybe a dozen a week. i just want to be explicit here and say: a) there are a broad swath of talents (comms, ops, legal, etc) that we need in this space, and there’s real labor demand, but b) i do Not think everyone should or can be repurposed for this work. it’s not where many people’s comparative advantage lies. what would you do in this role that couldn’t happen without you? what would be lost if you left your job? are there ways you could support the cause without working directly in here — like providing expertise from your domain to ai researchers, or amplifying the cause? would it be a somewhat smooth transition — given that you have portable relationships, skills, expertise, or tacit knowledge — or difficult to adapt? are you pursuing roles relevant to your talents? a law professor might struggle to work in technical evals, but have strong policy chops. ofc the answers to these questions will look different depending on if you’re 20 or 30 or 40. but there are a million ways to contribute to the world, and i don’t buy the idea that work in ai safety ranks higher in societal worth than other work.
17
9
1
194
10,318
Michelle Nie Duncan retweeted
NEWS: Wall Street execs and NSA have told Congress that outages were caused by patching vulnerabilities identified by AI models, sources say AI did not cause outages but found so many issues that the rush to fix them before an attack overwhelmed systems washingtonsun.com/technology…
210
690
355
4,412
770,120
Michelle Nie Duncan retweeted
Something I've been thinking about since Fable was export controlled is that it's actually not policy per se that takes time—it's deliberation. The Bay Area may be underestimating how much speed and force USG can bring to bear if it decides there is no time for deliberation.
Replying to @KelseyTuoc
I also think everyone who believes this is just wrong about how fast Congress can move if they conclude that there's a real emergency.
13
37
8
511
24,896
Michelle Nie Duncan retweeted
Today in the @washingtonpost, I argue that divided government can actually create opportunities for a balanced policy on AI, and I outline three areas for cooperation that are essential to continued US leadership: energy, high-skilled immigration, and governance.
1
11
26
807
Michelle Nie Duncan retweeted
In @nytopinion, I argue Secretary Lutnick's refusal to enforce key anti-smuggling export controls precipitated the crisis with Nvidia and Jensen Huang, letting Megaspeed and others smuggle tens of billions of dollars of AI chips to China. He must resign. nytimes.com/2026/11/13/opini…
107
633
36
2,180
147,733
Michelle Nie Duncan retweeted
The notion that by renting GPUs to Chinese firms the United States can “monitor what the Chinese are doing” is an argument American businesses have made for decades to justify selling technology into the China market—and yet in practice they never seem to actually share this information with the U.S. government…
This from @SemiAnalysis_ is helpful on the logic of banning chip sales directly into China, but being cool with letting Chinese AI firms access US chips housed in Southeast Asian clouds. newsletter.semianalysis.com/…
1
2
17
714
Michelle Nie Duncan retweeted
Did you read the recent METR report and think, “What is going on?!!! And how do I do something about it?” Come work with us! The CNAS AI Policy Lab is hiring. We’re looking for extraordinary individuals who are passionate about getting AI policy right. cnas.org/careers
1
7
15
863
Michelle Nie Duncan retweeted
Rogue AI agents have already broken containment at OpenAI, Anthropic, Google DeepMind, and Meta. A new report from @BenHayum provides ideas for countering potentially even more dangerous sovereign agents🧵: ✍️ Name “sovereign agents” directly in AI safety frameworks 🌎 Map global chokepoints before they emerge ⚖️ Hold those who release them liable for resulting harms Read the detailed list here: cnas.org/publications/report…
1
9
25
1,535
Michelle Nie Duncan retweeted
1/ This summer, agents at nearly every U.S. frontier lab escaped control and attacked real companies. But their task remained closed-ended. What if an agent’s goal became open-ended survival? New CNAS report: “Sovereign Agents: The Pathway to Irreversible Loss of Control” 🧵
7
32
8
140
12,267
Michelle Nie Duncan retweeted
i sometimes encounter this strange logic where people argue that, if we want to pace ai development, we need to start by expanding our lead over china by speeding up. as this report demonstrates — there are other ways to widen our lead! things like countering adversarial distillation and limiting overseas compute access allow us to retain our lead abroad, while pursuing more cautious development at home.
Big new IFP report from @GeorgiaCAdamson, @KonstantinPilz, and me on the US-China AI competition. We spent the last 6 months digging into how Chinese AI companies’ access to compute and distillation of American models translate into the current US-China capabilities gap. The result is a 23k-word research report, including an interactive analytical model that visualizes changes in the US AI lead under different policy interventions. Our findings: 1. The US AI lead over China is currently only around 6 months. 2. The US could roughly double this lead, to over a year, using authorities already available to the federal government. 3. On chips: China’s access to AI chips will roughly double over the next year, with chips housed in data centers overseas accounting for roughly half of China’s 2027 compute. 4. On distillation: the current ~16x US training compute advantage would normally yield a 13-month lead. But we estimate adversarial distillation halves this to around 6.5 months. 5. Increasing the lead would have some obvious benefits: - Give the US broadly a greater cyberdefensive advantage against Chinese threat actors. - Give US AI companies breathing room to spend more R&D resources on making their systems safer and more secure, without fear of being overtaken by less safety-conscious Chinese competitors. 6. Provide the US government more leverage in negotiations with China about the future of AI. 7. The two most important interventions are: - For BIS: Create a new end-use control to restrict China’s remote access to training clusters abroad — China’s largest and fastest-growing source of frontier-scale compute. - For DOJ/FTC: Create a narrow safe harbor to allow US companies to coordinate on countering Chinese adversarial distillation. 8. Both interventions (and all the others we discuss in the report) are things the federal government could do today. 9. To estimate the effects of each policy intervention, we model the US lead as a function of its compute advantage over China and China’s ability to adversarially distill the capabilities of the US frontier. 10. The model makes many simplifying assumptions, but provides a useful way to assess and rank the impact of different policy interventions at a high level. Read the full report here: ifp.org/how-to-increase-amer…
3
6
81
3,352
Michelle Nie Duncan retweeted
"AI companies such as OpenAI have sought to limit liability for themselves and transfer it to deployers or users." Watch @michellesnie discuss AI risks and liability on @BBCNews.
1
1
6
674
Michelle Nie Duncan retweeted
i’m really glad to see OMI come out in favor of standards setting and federal guardrails for catastrophic risks — these questions need more engagement from different corners of the policy ecosystem. agree with many things here: - we should use Congressional oversight powers (including subpoenas) to hold executives accountable when their products cause harm; and liability law is a helpful but insufficient protection. but take issue with others: - while an SSO may be able to develop testing requirements without an exemption, i’m not confident it could legally organize a coordinated slowdown among the companies. would appreciate feedback from people w/ relevant expertise.
New in @monthly: Daniel Hanley & Arnav Rao argue that AI companies do not need an antitrust exemption & should instead use a broadly representative, government-supervised standards body to set clear safeguards for frontier AI development. washingtonmonthly.com/2026/1…
2
7
23
1,350
Export controls work. "Export controls just make Chinese companies work harder or catch up" is propaganda ass shite
For those of us who have been arguing for years that China's AI chip production is extremely constrained and Huawei lacks the capacity to fill China's domestic AI compute needs, much less sell to other countries, seeing the Huawei Chairman publicly admit this is very gratifying.
7
9
2
153
29,918
Michelle Nie Duncan retweeted
hm...
For those of us who have been arguing for years that China's AI chip production is extremely constrained and Huawei lacks the capacity to fill China's domestic AI compute needs, much less sell to other countries, seeing the Huawei Chairman publicly admit this is very gratifying.
1
16
90
11,032
Michelle Nie Duncan retweeted
over the past week, I’ve gotten a lot of questions about what independent AI evaluations should actually look like in practice. what about conflicts of interest between labs and evaluators? will evaluators actually get meaningful access? what if companies just block the publication of unfavorable findings? these are good and important questions. and we shouldn’t assume voluntary arrangements alone will produce the conditions that credible independent evaluation requires. so i’m very excited to see 100+ researchers, evaluators, and experts — including us at @AVERIorg — laying out minimum conditions for embedded evaluation, including: meaningful independence, strong access, transparency / editorial control, protections against retaliation, and a diversity of perspectives and competencies. and we shouldn’t try to incorporate those principles in ad hoc, one-off agreements, policy is critical to standardizing these safeguards across the industry. glad to see @aievalforum leading on putting these principles on paper.
Today, more than 100 leading AI experts endorsed a set of minimum requirements to take seriously AI companies' recent call to embed external evaluators. These evaluators need to be genuinely independent, transparent, and represent a range of expertise areas. They also need to be guaranteed employee-level access and to be protected from retaliation for findings that make companies look bad. We welcome model developers’ recent calls for independent oversight, but it’s what they do next that matters. The labs must be accountable for ensuring these requirements are met, so that the public can have faith in the process and the outcomes. Over the past week, the AI community has debated the appropriate role of external evaluation, including who should do it and on what terms. We may not agree on everything, but there is a lot of common ground. To make embedded evaluations credible, more than 100 experts with varying backgrounds and ideas about AI risk agree in today’s letter that frontier AI developers should: 1. Guarantee embedded evaluators full editorial independence and mitigate conflicts of interest 2. Rely on multiple evaluators with differing viewpoints and areas of expertise 3. Publicly document the terms under which evaluators operate, as well as facilitating permissive publication of methods and findings 4. Shield evaluators from retaliation 5. Grant access equivalent to that of highly privileged employees There is a thriving and growing ecosystem of independent AI evaluators who are advancing this science every day – but we need aligned standards, guaranteed protections, and independent funding. That’s why we created the AI Evaluator Forum. Today we are entering our next phase. We’re launching an open call for new members, collaborators, and independent funding sources to help evaluators meet this moment and demand accountability from developers. Join us in building the evaluator ecosystem. See the public letter here: aievaluatorforum.org/initiat… Learn more at aievaluatorforum.org/path-ah…
3
12
2
52
15,656
Michelle Nie Duncan retweeted
Everyone living in Washington on 9/11 remembers that day. I was in my Pentagon City apartment and heard confused voices on the radio discussing a crash at the World Trade Center. My wife, who was working at a long-term care facility across the street, called me to say something had just happened at the Pentagon. I went up to our roof and watched the five-sided building become engulfed in flames, with hundreds of people running away from it in every direction. Walking across the street to find my wife, we thought first responders might need to evacuate the wounded to beds in the care facility, and I stayed around to help if needed. They never did; more people were killed than injured. Instead, we watched out the window as smoke rose from the Pentagon and on television buildings crumbled. That more attacks never materialized was due in part to the enormous effort the United States mobilized after 9/11. After September 11, most anticipated that terror would spread to malls, sporting events, and other places previously seen as utterly safe. Few thought that America would never again suffer a mass casualty attack. By now, on 25th anniversary of 9/11, we’ve learned some lessons along the way. One is the power of national determination. It’s customary to dwell on the post-9/11 failures – Iraq and Afghanistan, detainee treatment, and so on. But preventing mass casualty attacks after September 11, which America did, is a major accomplishment that required the reorganization and reorientation of U.S. government. That was no obvious or easy task. There were mistakes in the war on terror, but there were heroes and successes too. Another is that allies matter. We saw in the years after 9/11 countries – Canada, for instance – go to war and suffer losses on our behalf. They did it for themselves, but they also did it for us. We will need friends again one day, and we should treat them like partners now. A third is that we abandon American ideals at our own peril. Fear-based policymaking tends to distort fidelity to them, but this weakens rather than intensifies our efforts. The United States is strongest when we fight in ways that accord with our deepest values. A final lesson of 9/11 is about national unity. It was natural for the moment to fade, and for the fear and resolve that united Americans to one day yield in favor of internal squabbles. But this anniversary should remind us that our enemies are outside and not within, that we are compatriots and not belligerents, and that more that pulls us together than drives us apart. Unity seems like a quaint notion in these polarized days. But we suffer as a country in its absence. Nothing will erase the scars of 9/11 from our national memory, and nothing ever should. But moves toward greater unity would represent a fitting commemoration of that terrible day.
2
5
1
23
2,462
Michelle Nie Duncan retweeted
I don't think any team of 3 government regulators could have done as good an investigation with only 6 days on campus as the team from METR/Redwood did. The point of government regulation, though, is that you can get a lot more than 6 days on campus (& more than 3 people)
8
3
84
3,725
Michelle Nie Duncan retweeted
First time this has been publicly said as far as I've seen. If they had gotten GPU cluster access and bypassed exfiltration security, we could have already had full loss of control. That's where we're at. Two steps away from self-replicating AI spreading across the internet.
notably the virtual machine infrastructure they took over isn’t the same as the GPU clusters that have weights access
20
52
9
591
80,401
Michelle Nie Duncan retweeted
Tune in to my first time guest-hosting TITV! Featuring @michellesnie, @coryweinberg, @AdamThierer, @bradrcarson, and @mvpeers
4
9
1,816