whatimreading

Thursday, September 10, 2026 · 151 newsletters

Oil Hits $100, War Widens

Pulled from 197 newsletters after filtering notifications and receipts. Here's the signal cut from the noise, organized by trend.

Markets & War: Oil Crosses $100 and Nobody Blinks

The dominant story of the day is one that started as a market signal and became a geopolitical one. Brent crude hit $100 a barrel for the first time since July after US forces destroyed five Iranian tankers near Kharg Island, and Iran answered with a retaliatory wave, striking 10 vessels near the Strait of Hormuz and firing ballistic missiles at a US base in Jordan, according to Foreign Policy's World Brief. By evening, Bloomberg was reporting Brent had climbed past $101 as Chinese crude purchases picked back up too, a second demand-side push on top of the supply shock.

The financial-media reaction split into two registers. Snacks and EntryPoint treated it as a trading-desk story: stocks fell, chip names diverged, and semiconductor stocks actually rallied on a Qualcomm-Amazon chip deal even as the broader tape wobbled. Meanwhile Treasury Secretary Scott Bessent spent the day daring traders to bet against the yen, telling a Texas audience "I am the house now." Matt Levine's Money Stuff used the moment to needle at a deeper theory of how markets price correlated versus idiosyncratic risk, useful background for why a war in the Gulf moves bond yields in Tokyo.

The Canada front escalated in parallel. Trump moved to ban Canadian imports and bar Canadian firms from federal contracts, prompting Canada's retaliatory tariffs on up to $27.6 billion of US goods. Paul Krugman called the entire dispute "deranged Trump syndrome" dressed up as trade policy, noting Canada actually ran a near-balanced overall trade relationship with the US last year. The EU, watching the same spectacle, is now reportedly angling for a closer relationship with Canada than with the UK.

None of this is separable. War premium in oil, a manufactured trade war with the closest US ally, and a Treasury Secretary publicly goading currency traders are three symptoms of the same thing: an administration treating volatility as a tool rather than a risk to manage. Markets are pricing that in slowly, and not fully.

AI: The Week Two Anthropic Researchers Said the Quiet Part Out Loud

The extinction-risk conversation stopped being theoretical. Jacob Coxon, who worked on pretraining at both OpenAI and Anthropic, resigned this week and posted that both companies are "racing straight to self-improving superintelligence and gambling with our lives," a thread Semafor Flagship reported passed 100 million views. What made it land, per Guillermo Flor's AI Market Fit, is that Anthropic's own alignment lead Evan Hubinger publicly agreed on the record: "we really do earnestly believe AI could kill all humans... I personally think it is >10% within the next decade." Ted Gioia called it the opening scene of a horror movie, and even Bridgewater's Greg Jensen and Nir Bar Dea, republished by John Ellis's News Items, framed AI governance as "the most important policy decision of our lifetime."

Meanwhile the product cycle didn't pause for the apocalypse talk. Meta launched Muse, its first consumer AI agent, which Stratechery's Ben Thompson argued may matter more to daily life than OpenAI's math breakthrough, precisely because it's mundane: booking travel, sending emails, shopping through Stripe's Link. Linas's Fintech Pulse read Muse as a bet that trust, not model quality, wins the agent race against Grok and Gemini, a bet Meta is making from a weakened trust position just weeks after settling child-safety claims for up to $18 billion. Axios AI+ and The Information both flagged that Muse runs in a sandboxed VM with a separate "Sentinel" approval agent, security theater aimed squarely at making people forget the source.

And the OpenAI math claim got messier than the headlines suggested. OpenAI said its models solved the Navier-Stokes Millennium Prize problem, but The Information's AI Agenda reported the achievement is shadowed by questions over whether OpenAI's models learned from two human mathematicians' unpublished work on the same problem via Codex, the exact kind of "learning from customer usage" concern that undermines trust in frontier labs generally. TLDR and AlphaSignal both covered the 10,000-agent, 88-hour approach uncritically; only the trade press caught the credit dispute.

The split here isn't between optimists and doomers, it's between people paid to ship product this quarter and people whose job is to think past it. Both groups are now saying alarming things in public at the same time, which is new.

Trade & Industrial Policy: Chip Deals Are the Real Arms Race

Qualcomm's pact to supply Amazon with custom AI chips sent semiconductor stocks up more than 1% even as the broader market fell, EntryPoint reported, part of what Alex Wilhelm at Cautious Optimism called AI capability becoming correlated with balance-sheet risk. Paul Kedrosky's roundup made the sharper point: rating agencies are starting to scrutinize hyperscaler debt and take-or-pay compute contracts that are about to start billing, meaning the financing side of the AI boom is catching up to the spending side. Separately, The Information's China digest reported DeepSeek's revenue jumped tenfold and MiniMax hit an $800 million annualized run rate, evidence that US chip restrictions never actually slowed China's AI economy, just forced workarounds that are now being exported.

The chip fight and the AI-extinction fight are the same fight from two angles: one side is arguing about who profits from the buildout, the other about whether the buildout should exist at all. Nobody running earnings calls this week engaged with the second question.

Politics: The Convention Is the Distraction It Looks Like

Republicans opened a two-day midterm convention in Dallas that Trump personally requested, branded internally as "Trump-a-palooza." Dan Pfeiffer at The Message Box argued the entire spectacle, including hagiographic PAC ads indistinguishable from 2024 campaign material, is a strategic error dressed as strength. Bloomberg's Nancy Cook covered the same event as an unprecedented gamble, noting Trump is centering himself in a midterm at the exact moment his approval sits at a record low. Bloomberg Opinion's Jonathan Levin was blunter: Trumpapalooza isn't going to rescue your retirement, pointing to Medicare's looming fiscal burden as the thing actually deciding markets.

The convention is also covering for genuine dysfunction. Semafor DC noted Senate Majority Leader John Thune is conspicuously absent from the lineup, and Popular Information's Judd Legum used the 9/11 anniversary to fact-check Trump's decades-old claim that he personally led a construction crew to Ground Zero, a story Legum notes has no supporting evidence and contradicts records showing Trump's own company employed roughly a dozen people at the time.

Redistricting and voting rights stayed volatile. Marc Elias reported the Supreme Court rejected Missouri's gerrymandered map gambit, only for a Republican Secretary of State to say he'd proceed with it anyway, prompting the state's Chief Justice to demand he show cause. Democracy Docket separately flagged that Tea Party Patriots is training poll watchers in battleground states ahead of the midterms, and the Trump administration asked the Supreme Court for the second time in three days to unblock a Social Security-linked voter database a judge ruled had already wrongly purged citizens.

Trump's approval is underwater, his party is throwing him a party anyway, and the actual mechanics of the midterms (poll watchers, gerrymandered maps, voter databases) are being contested court by court while the cameras point at Dallas. That's not a coincidence of scheduling.

Ideas Worth Reading

Outside Interests

Data Worth Noting

Three Takeaways for You

The oil spike is doing more work in this news cycle than any single data release: it's connecting a shadow war in the Persian Gulf to Bessent's currency posturing to a Canada trade fight that Krugman calls gratuitous. When three unrelated newsletters (Bloomberg's markets desk, Foreign Policy's world brief, and a Substack fintech writer) all converge on the same $100 number within hours, that's not noise, that's the market finally pricing in that this administration treats escalation as a feature.

The AI safety conversation crossed a threshold this week that the product-launch conversation didn't notice happened. Two Anthropic researchers put a public number on human extinction risk in the same 48 hours that Meta shipped a shopping agent and OpenAI disputed credit for a math proof. Nobody in the room building the products is treating the researchers' number as an input to their roadmap, which is itself the story.

If you only read three pieces, make it Ted Gioia's The Last 24 Hours Are the Opening Scene in a Horror Movie for the AI safety cascade, Paul Krugman's Adding Injury to Insult, Canada Edition for the trade war's actual stakes, and Judd Legum's Trump's 9/11 Fantasies for a reminder that the small lies are still worth checking even in a week this loud.