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…

NEWS: 19 FORMER ECONOMIC AND NATIONAL SECURITY LEADERS ENDORSE THE BIPARTISAN PERMITTING REFORM BILL Today, a group of former Cabinet members and White House principals across seven Democratic and Republican administrations sent a letter calling on Congress to pass BAAJA. Here are the signatories: Janet YELLEN, Michael R. POMPEO, Steven T. MNUCHIN, Leon E. PANETTA, Timothy F. GEITHNER, Henry M. PAULSON, Jr., Robert E. RUBIN, Stephen MIRAN, Jared BERNSTEIN, Tyler GOODSPEED, Jason FURMAN, Glenn HUBBARD, Laura D’Andrea TYSON, Michael J. BOSKIN, Jeffrey ZIENTS, Jake SULLIVAN, Brian DEESE, Larry KUDLOW, Allan B. HUBBARD Key highlights from the letter: 1. “We are not building infrastructure fast enough to remain globally competitive… The median time for a new power plant to move from an interconnection request to commercial operation has more than doubled, from less than two years for projects built between 2000 and 2007 to more than four years for projects built between 2018 and 2024. 2. China has built more high-voltage transmission in the last fifteen years than the U.S. has built in all of its history... In 2024, China accounted for 31.5% of global knowledge- and technology-intensive manufacturing value added, compared with 20.7% for the United States. 3. Federally funded or permitted projects require NEPA review, but private projects often do not, creating a threat of failure precisely where the government is trying to be proactive, strategic, and engaged. 4. A project that spends years awaiting approval must continue carrying development and financing costs while remaining exposed to rising input costs. The IEA reports that prices for utility-scale transformers increased 26% in 2024, while lead times for some transformers have stretched as long as six years. 5. Mining projects also face tremendous uncertainty: a 2026 study by researchers at the Colorado School of Mines found that permitting for modern mining projects takes an average of 5.5 years, rising to 8.9 years for contested projects. The U.S. currently has the second-longest mine development timeline in the world. 6. The Bipartisan American Affordability and Jobs Act is the most significant bipartisan legislative effort that Congress has undertaken in decades to tackle these challenges. 7. BAAJA addresses key constraints stifling American industry by requiring interregional transmission planning, strengthening the federal backstop for nationally significant transmission lines, and creating a faster, more predictable process for connecting new generation to the grid. 8. BAAJA also makes permits more durable by imposing tighter timelines for litigation and limiting frivolous lawsuits that can halt projects indefinitely, while preserving community input and judicial review of substantive agency decisions.” The letter was organized by @arnabdatta321 on behalf of @IFP and covered exclusively by the WSJ this morning.

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…

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…

I must be among an extremely small group of people (n=1?) that have both 1) trained a frontier LLM and 2) designed and synthesized custom viruses in a lab with my own two hands. And I think that the takes on AI killing us all by creating dangerous viruses is total bogus.

labs now commit to audits of the world’s most powerful models. who trains the auditors? which university is gonna be the first in the world to launch a Frontier AI Auditing program? would be a pretty sweet claim to fame wouldn’t it dm let’s talk

JUST IN: D.C. Circuit upholds the Trump administraiton's decision to ice out Anthropic/Claude, finding the government's claim of security risks to have "ample support." documentcloud.org/documents/…

Third-party evaluators mostly test frontier AI models through an API. That misses the risks that come from how companies build and use AI internally. Our new paper makes the case for embedded assessments. It was led by Jacob Charnock. The other co-authors are Sophie Williams, @zaheedkara, @Manderljung, Alejandro Tlaie Boria, @StephenLCasper, @AnkaReuel, and me. 🔗 Read the paper: governance.ai/research-paper…

I cannot impart on progressive folks more strongly that it is not just Sam Altman sounding these alarms. I also do not care for or trust Sam Altman! But listen to the people actually building this stuff who are freaked out. Maybe they are wrong but they are not part of some profit motivated psyop.

Plaintiffs suing the leading AI companies over "pacing" argue purported agreement to reduce the rate at which their AI improves violates antitrust law. "AI alone could add trillions annually to the global economy... the product of the competitive race Defendants' agreement now threatens to slow."