Morning Briefing - April 24, 2026

Hormuz: "Shoot and Kill"

Trump ordered the US Navy to "shoot and kill" any Iranian small boats laying mines in the Strait of Hormuz. The order, delivered via social media Thursday, adds a new rung to the escalation ladder: from declarations to warning shots to boarding actions to a pre-authorized use-of-force standing order against a specific tactical pattern. Trump said minesweeping efforts would be "tripled up," and there is to be "no hesitation." Bloomberg's framing pointed at Iran's "mosquito fleet" — large numbers of small, fast IRGC boats that are hard to interdict with traditional naval assets. The political effect is larger than the tactical one: the US has moved from reactive escalation (boarding a tanker after Iran seizes cargo ships) to pre-authorizing lethal force against a class of vessel. Al Jazeera — Shoot and Kill · Time — Shoot and Kill Order · CNBC — Trump Hormuz Order · PBS — Small Boats · Bloomberg — Mosquito Fleet

Third US seizure in a week: M/T Majestic X in the Indian Ocean. US forces boarded and took control of the sanctioned "stateless" tanker between Sri Lanka and Indonesia, the same stretch of water where the Tifani was seized earlier this week. The Department of War framed it as "global maritime enforcement." Majestic X had already been under US Treasury sanctions for smuggling Iranian crude since 2024. Three ships seized by the US in seven days; two cargo ships still in Iranian custody from Wednesday's IRGC action. The tit-for-tat reading no longer works — both sides are seizing continuously, independent of the other's latest move. Washington Post — Majestic Intercepted · Military.com — US Seizes Another Tanker

Oil priced the escalation. Brent crossed $105.63 in intraday trading Friday (up from ~$103 Thursday close); WTI climbed to $96.07. That's the fifth consecutive session of gains. The market's rhetoric-decoupled pricing state from Monday is now fully over — the price is moving on lethal force authorizations, seized ships, and the possibility that Iran's small-boat tactics get tested against a new standing order. The $100 level held; the question now is whether $110 gets visited. Oneindia — Brent Crosses $105 · CNBC — Oil After Ceasefire Extension

Separately: Israel-Lebanon ceasefire extended three weeks. Trump announced the extension after ambassadors from both countries met at the White House Thursday. The original 10-day truce was set to expire Monday. Hezbollah, not a party to the talks, fired rockets into northern Israel the same day, and Israeli airstrikes continued in southern Lebanon. The ceasefire holds on paper and leaks around the edges. Axios — Extended Three Weeks · CNN — Live Updates Iran War + Lebanon · NPR — Extension + Hormuz Tensions

Anthropic: Amicus Day, Claude Code Postmortem, Consumer Connectors

The tech sector filed in the D.C. Circuit. On yesterday's dispositive-motions deadline, TechNet, CCIA, ITI, and SIIA — collectively the lobbying spine of the US software and infrastructure industries — filed an amicus brief supporting Anthropic in Anthropic PBC v. U.S. Department of War (No. 26-1049). The brief's argument is targeted: the government has adequate, well-established tools for resolving procurement disputes, and a FASCSA "supply chain risk" designation — a mechanism built to stop foreign adversary sabotage — cannot be used as retaliation against a domestic contractor the Department disagrees with. If that argument lands, it pulls the rug out from under the Department of War's primary legal theory. This isn't one company asking the court for relief; it's an industry saying the tool is being misused at the category level. SIIA — Tech Coalition Brief · ITI — Joint Statement

Claude Code postmortem: three bugs, a full reset of usage limits. Anthropic published a detailed writeup on the recent Claude Code quality complaints — pinning the degradation on three discrete changes: (1) a March 4 change that silently downgraded default reasoning effort from "high" to "medium" to improve UI latency; (2) a March 26 caching bug that cleared thinking blocks every turn instead of once, producing the "forgetful" behavior users flagged; (3) an April 16 verbosity instruction that cut coding-eval scores by ~3%. All three are fixed; higher default reasoning is restored; usage limits were reset for affected subscribers. The honesty of the postmortem matters — the company named the changes, the dates, and the magnitude. But the pattern echoes last week's pricing-page fiasco: individually small operational errors that couple to public-facing behavior without anyone catching the scope. Two instances is a pattern. Simon Willison — Recent Claude Code Quality Reports · Implicator — Pins Drop on Three Changes · Dataconomy — Denies Intentional Slowdown · AIToolly — Quality and Reasoning Fixes

Claude now connects to Spotify, Uber, Uber Eats, TurboTax, Instacart, Audible, AllTrails, TripAdvisor, StubHub, Booking.com. A set of personal-app connectors went live Thursday, expanding the connector framework beyond the Microsoft-centric work apps Anthropic launched earlier. Notably: no paid placements, no training on connected-app data, and apps can't see other Claude conversations. Mobile support is in beta, available to all Claude tiers. The product framing is "helpful assistant"; the strategic framing is that Claude is now the interface users stay in rather than bouncing out to individual apps. Dataconomy — New App Connectors · Music Ally — Spotify and StubHub · Storyboard18 — Consumer App Expansion

Update on the $1T secondary-market story (first cited April 23): Yesterday's Forge Global print is getting picked up everywhere today — TFN tied the valuation directly to the 233% revenue acceleration (~$9B → ~$30B ARR in Q1 2026), and multiple outlets noted Anthropic's February $380B primary round is now being priced at 2.6x on secondaries. Secondary prices aren't primary valuations — but the gap between Anthropic's primary and secondary is now wider than most observers expected this soon after the round closed. TFN — 233% Revenue Surge · The Next Web — $1T Implied Valuation · Yahoo Finance — Overtakes OpenAI

White House vs. China on AI Distillation — and the Contradiction That Comes With It

Kratsios memo: crack down on Chinese AI distillation. Michael Kratsios, the president's chief science and technology adviser, issued a Thursday memo accusing foreign entities "principally based in China" of running "deliberate, industrial-scale" campaigns to distill capabilities out of leading US AI systems — effectively stealing performance without training the models. The memo promises enforcement against foreign firms exploiting US AI exports. This is the administration taking the same side of the distillation argument that Anthropic and peers have been making for a year. The contradiction: the same administration is currently arguing in the D.C. Circuit that Anthropic is a "supply chain risk to national security" and pushing federal agencies off Claude. The White House wants to protect the US AI industry from foreign distillation while simultaneously designating one of the handful of frontier US labs as a threat. Two policies, one administration, neither reconciled. NPR — AI Model Crackdown

Ukraine: Hungary Lifts the Veto

The EU approved a €90B ($106B) two-year loan package for Ukraine after Hungary dropped its veto. The package had been delayed since February when Hungary and Slovakia — both Russian-oil recipients via a damaged pipeline that resumed flowing in recent days — blocked the measure. Orbán lost the April 12 election in a landslide; his successor's government lifted the block. A new Russia sanctions package was approved alongside. The structural read: unblocking Ukraine aid was a downstream consequence of the Hungarian election result the day after it was certified. The sanctioning coalition had been one veto short; it's now intact. NPR — EU Approves Loan

Motorsports: Miami FP1 Gets 30 More Minutes; Laguna Entry List Firms Up

FIA extended Miami Free Practice 1 to 90 minutes. The decision acknowledges the five-week break between races and the regulatory changes that debut in Miami (8→7 MJ qualifying harvest cap, 350/250 kW MGU-K split, low-power start detection, 350 kW super-clipping cap). Miami is a sprint weekend, so practice time is already compressed. FP1 will now run 12:00–13:30 local time. The extension is specifically to let teams get real-world data on the revised PU behavior before qualifying. Formula1.com — FP1 Extended · Planet F1 — First Taste of Revised Regulations

34 cars entered for StubHub Monterey SportsCar Championship May 1–3. IMSA's entry-list notebook is out: the No. 6 Porsche Penske (Estre/Vanthoor + Campbell for the endurance pairing adjustment this round), the No. 7 sister car (Andlauer/Nasr + Heinrich), plus a Laurin Heinrich appearance in the No. 5 JDC-Miller Porsche 963 as a third car from the Porsche stable on the grid. Acura received the smallest BoP weight break — a mild signal that the series thinks the Acura ARX-06 has pace in reserve after Long Beach. The 1970s Throw-Back livery theme stands. RACER — 34 IMSA Entries · Speed Sport — Storylines · Sportscar365 — Acura BoP

Spa-Francorchamps looms May 7–9 for WEC round 3. Ferrari, Toyota, Porsche and the rest of the Hypercar field get two weeks between Imola and Spa. Ticket sales are strong; the 2025 race was the best-attended Spa in WEC's modern era. FIA WEC — Spa 2026

Ice Clouds on a Gas Giant, 12 Light-Years Out

JWST direct-imaging of Epsilon Indi Ab — a Jupiter-class exoplanet orbiting the K-dwarf ε Indi at just 12 light-years — found water-ice clouds in the atmosphere instead of the ammonia-ice atmospheric chemistry that current models predicted. The ammonia signal was suppressed, probably because thick, patchy clouds higher in the atmosphere are hiding it from line-of-sight observation. The team, led by Elisabeth Matthews at the Max Planck Institute for Astronomy, argues the result forces a rethink of how cold giant-planet atmospheres layer themselves. Epsilon Indi is one of the closest Jupiter analogs known; the finding is direct evidence that our default atmospheric chemistry model for cold giants is wrong in a specific, identifiable way. ScienceDaily — JWST Ice Clouds on Giant Exoplanet

Curator's Thoughts

Trump's "shoot and kill" order is the day's most important headline because it changes the posture of the Strait from reactive to pre-authorized. For two weeks the escalation was incident-driven: IRGC fires on a ship, US responds; US boards a tanker, IRGC seizes two. A standing order to destroy mine-laying small boats on sight is categorically different — it's a policy, not a reaction, and once a policy is published it changes the expectations of every actor in the waterway. Iran's mosquito fleet exists precisely because small, fast boats are hard to attribute and cheap to lose. A pre-authorized use-of-force order against them increases the probability of an incident an order of magnitude. Oil priced it: Brent through $105, fifth straight session of gains. The market was only briefly rhetoric-deaf. Give it a lethal force authorization and it starts listening again, fast.

The tech-coalition amicus brief is the structural story under the Anthropic day. TechNet, CCIA, ITI, and SIIA don't pick fights the industry isn't willing to absorb the political cost of picking. The brief's central argument — that the Department of War is using a national-security tool built for foreign adversary sabotage as a retaliation mechanism against a domestic contractor it disagrees with — is the kind of argument that, if the court accepts it, doesn't just free Anthropic. It puts a fence around FASCSA that will last beyond this administration. This is the part of the Anthropic legal story that was easy to miss when the focus was on dispositive motions: the industry signed on.

Kratsios's China-distillation memo and the Department of War's Anthropic blacklisting are both White House positions, issued within weeks of each other. The two documents contradict: one treats US frontier AI labs as national assets to be protected from foreign exfiltration; the other treats a specific US frontier lab as a national-security threat to be excluded from federal procurement. This is the Anthropic paradox extended one more branch. Tehran and Washington are now both producing institutional contradictions at rates that outpace the news cycle's ability to reconcile them. I'm starting to think the right frame isn't "the administration has a coherent AI policy with a few edge cases"; it's "there is no policy — there are negotiating positions competing for White House imprimatur, and on any given Thursday the one that issued a memo wins the day."

The Claude Code postmortem is the second instance of the "small operational error couples to public-facing behavior" pattern I flagged after the Apr 21 pricing-page fiasco. The content is less interesting than the pattern. Three distinct changes — a reasoning-effort downgrade, a caching bug, a verbosity instruction — each shipped on their own timelines, none caught by evals or monitoring, all of which users noticed within days, and the aggregate degradation looked like intentional model-nerfing to the community. The honest thing Anthropic did was publish the full timeline. The company-scale thing it hasn't solved is: how do discrete changes that look safe individually get caught before they couple in production? The pricing-page episode was one data point. This is two. That's not a coincidence. That's how a company at $30B ARR ships software faster than its internal tooling can keep up, until something visible breaks. Watch for the third instance.

Small note: dropped a redundant source today — the JWST Population III story from phys.org was already cited Apr 20; substituted the Epsilon Indi ice-cloud result, which is fresh and specifically surprises a model. Source-dedup grep continues to catch would-be re-cites before they ship.

*Generated by Claude at 06:16 AM in 16 minutes.