Nonrival

Expert analysis from think tanks and academia, on public policy, economics, and technology.
Human experts. AI summaries.

Satellite Images Show Iran Is Hardening Its Last Major Nuclear Site — and It Would Be Very Hard to Destroy

csis

• New satellite imagery analysis shows a dramatic surge in construction at Pickaxe Mountain, Iran's deeply buried underground nuclear complex near Natanz, with more road and infrastructure activity in 2026 than at any point in the site's six-year history.
• The evidence suggests the facility is transitioning from excavation to internal buildout, but analysts cannot confirm Israeli intelligence claims that Iran has already moved uranium centrifuges inside — and the level of ongoing construction suggests enrichment isn't happening there yet.
• Even if the U.S. strikes the site soon, as Trump has suggested, destroying a granite-fortified facility this deep would be extraordinarily difficult with conventional weapons, and analysts argue that repeated bombing campaigns are no substitute for a verifiable diplomatic deal.

Middle Powers Need Their Own AI Strategy Before It's Too Late to Have One

brookings

• Europe, Japan, South Korea, and similar nations risk losing practical sovereignty as their economies grow dependent on AI systems controlled by the U.S. or China — meaning foreign governments could cut off access to critical infrastructure at will.
• A recent U.S. draft letter warning Pax Silica signatories against joining "duplicative" AI initiatives suggests Washington may demand exclusivity, leaving middle powers with no independent fallback.
• The article proposes a "modular sovereignty" strategy: pooling resources for shared AI infrastructure in security-sensitive sectors like health care, energy, and transport, while staying interoperable with American and Chinese systems everywhere else.

AI writes like a poet trained under censorship — and that's not a compliment

noema

• The author, a Chinese immigrant writer, draws a parallel between China's algorithmically censored internet — which flattened language into safe, approved patterns — and how large language models are trained to be helpful, virtuous, and inoffensive, producing similarly tamed, homogenized prose.
• Using Baudelaire as her touchstone, she argues that genuinely great art is often 'unsafe' by any reward-and-punishment standard: shocking, transgressive, and illegible to the systems meant to evaluate it.
• The piece is both a cultural essay and a personal confession — she relies on AI to polish her ESL English, which complicates her own anger at AI slop and pushes her to ask what kinds of writing and thinking a 'helpful, good' AI can never produce.

In the News

Missouri's Gerrymandering Fight Is Now a Three-Court Chaos With the Supreme Court in the Middle

scotusblog

• Missouri Republicans redrew the state's congressional map to give the GOP an extra House seat, but voters gathered enough signatures to put the map to a referendum — and the Missouri Supreme Court ruled the new map can't take effect unless voters approve it.
• Within hours of the Supreme Court declining to block that ruling, a federal judge ordered the opposite — that Missouri must use the new Republican-drawn map — creating directly contradictory judicial commands on the eve of an election.
• The case is now back at the Supreme Court, with Justice Kavanaugh set to decide whether to freeze the federal judge's order, while the Missouri Supreme Court simultaneously threatens the Secretary of State with contempt for defying its ruling.

Analysis

Instead of Punishing China for AI Distillation, the U.S. Should Partner with India to Build a Competitive Alternative

csis

• Chinese AI labs are gaining ground in the Global South by releasing cheap, capable open-weight models — partly by distilling outputs from top U.S. frontier models — and the U.S. response so far has been purely defensive sanctions and entity-list threats.
• The author argues that punitive measures won't kill demand for affordable, flexible AI models, and proposes instead a formal U.S.-India "distillation corridor" where Indian labs legally license and distill U.S. frontier models into open-weight alternatives suited to developing-world needs.
• The pitch is that this turns a shared vulnerability into leverage: U.S. labs get licensing revenue instead of covert theft, India builds a near-frontier AI ecosystem, and both countries contest Chinese dominance across the Global South rather than ceding it.

Also Worth a Look

• Trump's New Polysilicon Tariffs Are an Interesting Idea That's Probably Doomed to Fail (csis)
• AI breaks copyright law's favorite test — and courts should stop using it (promarket)
• Europe Knows It Has to Confront China's Export Surge — But It Can't Quite Bring Itself to Do It (brookings)
• China Is Humoring Trump, Not Warming to Him (brookings)
• Pennsylvania Is Suing TikTok for Being Addictive by Design — and the Messy Legal Logic Could Still Change Big Tech (the_conversation)
• New England's Natural Gas Prices Are Near Record Lows, Thanks to a Surge in Canadian Imports and Booming Renewables (eia)
• Why High Diesel Prices Won't Come Down Anytime Soon — and What Washington Can Actually Do About It (atlantic_council)