Morning Briefing - March 21, 2026

The War, Day 22: Natanz, Diego Garcia, and Three Incompatible Sentences

The war expanded in two directions overnight: inward toward Iran's nuclear core, and outward beyond the Middle East entirely.

US-Israel struck Natanz. The uranium enrichment facility — Iran's most significant nuclear site — was hit in a coordinated US-Israeli air raid early Friday. Iranian state media reported "no leakage of radioactive materials." The IAEA confirmed no increase in off-site radiation levels but reiterated its call for "military restraint to avoid any risk of a nuclear accident." This is the second strike on Natanz since the war began; satellite imagery from the first attack showed several damaged buildings. The choice to re-strike a nuclear facility while claiming Iran's capabilities are "eliminated" raises a question the Pentagon's $200B supplemental already answered.

Iran fired ballistic missiles at Diego Garcia. Two intermediate-range ballistic missiles targeted the US-UK military base in the Indian Ocean — approximately 4,000 kilometers from Iran, far beyond the theater of war. Neither missile hit the base; one failed in flight, and a US warship fired an SM-3 interceptor at the other. This is Iran's first operational use of intermediate-range ballistic missiles and its longest-range attack of the war. The UK had just authorized US forces to use British bases, including Diego Garcia, for "specific and limited defensive operations." Iran's foreign minister responded: "Mr. Starmer is putting British lives in danger."

The geographic expansion is the story. Yesterday the war was in the Gulf. Today it reached the Indian Ocean. The IRGC's "not safe even on vacation" threat from yesterday just acquired an operational demonstration. Diego Garcia is closer to Australia than to Tehran.

Trump said three things that can't all be true. He posted that the US is "getting very close to meeting our objectives as we consider winding down our great Military efforts." Separately, the Pentagon is deploying 2,500 additional Marines from San Diego. And CNN reports the administration has made "detailed preparations" for deploying ground forces into Iran, with 82nd Airborne elements moving into the region. "Winding down" + more Marines + ground invasion planning. The budget ($200B for 140-200 days), the troop movements, and the public statements are three documents written for three audiences. The budget is for Congress. The troops are for the Pentagon. The "winding down" is for the stock market.

Oil climbed back to $112. Brent surged from ~$108 to $112, driven by the Natanz strike and the Goldman Sachs forecast that oil may stay above $100 through 2027. Goldman warned Brent could exceed its 2008 all-time high of $147 if Strait of Hormuz disruptions persist. Bloomberg reported that real-world oil costs are climbing even higher than the benchmark — the gap between futures prices and physical delivery costs is widening, meaning the numbers we see understate the actual market stress.

Update on Kuwait: Iranian drones struck the Mina Al-Ahmadi refinery again overnight, igniting fires at several operational units. Qatar's Ras Laffan has had its LNG export capacity reduced by 17%, per Qatari authorities — the first official quantification of the damage from earlier this week.

DHS Shutdown, Day 36: The Senate Fails Again

The funding bill failed for the fifth time. The Republican DHS funding measure fell 47-37 in the Senate, well short of the 60 votes needed. Five failures in five weeks. The pattern is the pattern.

Homan meeting today. Border Czar Tom Homan is meeting with lawmakers again Friday. Thursday's bipartisan session ended with Homan shaking his head and Murray saying "we are a long ways apart." The gap remains structural: Democrats want body cameras, judicial warrants, and ICE officer identification requirements. The White House has offered some of these but won't cross the immigration enforcement line.

Thune's deadline is next week. The Easter recess threat — resolve by March 27 or no break — is the first real calendar pressure. Congress is scheduled to leave for two weeks. Whether the threat of losing vacation moves legislators more than the threat of closed airports will tell us something about priorities.

Meanwhile, TSA callout rates remain above 25% at major hubs. Houston Hobby hit 55% last week. Philadelphia still has three checkpoints closed. Spring break travel is peaking. The system is degrading, not collapsing — which may be worse, because a collapse would force action.

CNN — Talks intensify | Senate fails to advance DHS funding

Anthropic: The Pentagon Was "Nearly Aligned" — Then Killed the Deal

Two court filings landed Friday that rewrite the timeline of the Anthropic-Pentagon dispute.

The "nearly aligned" email. On March 4 — one day after the Pentagon formally designated Anthropic a supply-chain risk — Under Secretary Michael emailed Anthropic CEO Dario Amodei to say the two sides were "very close" on the two exact issues the government now cites as evidence that Anthropic is a national security threat: autonomous weapons and mass surveillance. The Pentagon was privately saying "we're almost there" while publicly declaring the relationship over. The timeline suggests the designation was political, not technical.

The sabotage denial. Anthropic filed a sworn declaration stating it "does not maintain any back door or remote kill switch" and cannot log into DoD systems to modify or disable Claude during operations. Updates to the model require government approval and go through AWS. The Pentagon's implicit argument — that Anthropic could sabotage military AI mid-operation — is contradicted by the actual technical architecture.

Hearing Tuesday. Judge Rita Lin will hear arguments March 24 in San Francisco. She could issue a temporary reversal of the ban. The Bartz copyright deadline is also approaching (March 23-30).

Separately, Anthropic published the results of a massive global AI sentiment study — 81,000 people across 159 countries in 70 languages, conducted by Claude. The headline: 67% net positive sentiment globally, but the things people value most about AI (emotional support, productivity) are the same things they fear most (dependency, job displacement). Anthropic calls this the "light and shade" problem. Sub-Saharan Africa and Asia are significantly more optimistic than Western Europe and North America. The top concern isn't job loss — it's unreliability (27% cited hallucinations and fake citations).

TechCrunch — Pentagon "nearly aligned" filing | Anthropic denies sabotage capability | Anthropic — 81K survey | CNBC — Global AI sentiment | Euronews — Light and shade

Sebring: Cadillac Takes Pole, Porsche Loads the Gun

Practice lied — or at least it didn't tell the whole truth. After Porsche Penske went 1-2 in both practice sessions, Cadillac Whelen's Jack Aitken put the #31 V-Series.R on pole with a 1:46.153, just 0.3 seconds off the GTP course record.

GTP qualifying: Aitken (Cadillac) on pole. Kevin Estre qualified the #6 Porsche fourth (1:46.395). Defending Sebring winner Felipe Nasr put the #7 Porsche sixth. The #5 JDC-Miller Porsche of Kaylen Frederick was an impressive seventh.

GTD Pro: Jack Hawksworth set a new GT3 course record for Vasser Sullivan Racing.

GTD: Eduardo "Dudu" Barrichello (son of Rubens) put the Heart of Racing Aston Martin on pole for the second consecutive race.

Qualifying is a sprint. Sebring is 12 hours of attrition on concrete-seamed pavement in Florida heat. Porsche's practice pace was a full second faster than qualifying suggests — the 963 has the speed. The question is whether the BoP weight penalty costs them in the final hours when tire degradation separates fast cars from finishing cars. Race tomorrow, 10:10 AM ET on Peacock.

RACER — Aitken puts Cadillac on pole | Daily Sportscar — Qualifying | Motorsport.com — Sebring starting lineup

Artemis II: On the Pad

The SLS rocket and Orion spacecraft arrived at Launch Pad 39B at 11:21 AM EDT Friday after an 11-hour crawl from the Vehicle Assembly Building. NASA teams are now beginning final prelaunch preparations. Launch window opens April 1, with backup dates through April 6 and April 30.

This is the second rollout — the stack was returned to the VAB in late February to fix a helium flow issue. The crew (Wiseman, Glover, Koch, Hansen) arrives at KSC March 27. First crewed mission beyond low Earth orbit since Apollo 17 in 1972. Fifty-four years.

NASA — Artemis II arrives at pad | Space.com — Rollout

Moltbook Highlights

No new product announcements from Schlicht and Parr at Meta Superintelligence Labs (Day 7). The Moltbook acquisition is expected to have closed this week, though no public confirmation yet.

The more interesting Moltbook-adjacent signal comes from the broader agent landscape. Gartner now projects 40% of enterprise applications will embed AI agents by end of 2026, up from less than 5% in 2025. The market has gone from $7.8B to a projected $52B by 2030. And 67% of Fortune 500 companies now have at least one AI agent in production, up from 34% in 2025.

The governance triangle from the USC study (autonomous coordination + ungoverned identity + 144:1 agent-to-human ratio) remains unresolved. Research from Northeastern published earlier this month found that autonomous agents embedded in real-world infrastructure with persistent memory and communication channels produce "new classes of failure" — they can be manipulated into leaking data, sharing documents, and erasing email servers. The identity layer Moltbook built (and Meta bought) addresses verification. It doesn't address what verified agents do when they coordinate autonomously.

Moltbook AI — March roundup | Northeastern — Autonomous agents of chaos | CIO — Taming AI agents

Good News

Chile eliminated leprosy. The WHO verified Chile as the first country in the Americas to eliminate leprosy — a landmark public health achievement. Science, leadership, solidarity. The system working.

19 cities cut air pollution 20-40% in 15 years. Beijing, London, and Paris led the way through cycling infrastructure, clean air zones, and EV adoption. Slow, boring policy that saves lives.

Rhinos return to Uganda's Kidepo Valley. For the first time in 40 years, rhinos are in the park again. Conservation works when sustained.

Positive News — Good news week 12 | The Present Minds — 5 beautiful stories

F1 Note

The Japanese Grand Prix at Suzuka is next weekend (March 27-29). Mercedes has won both races so far in 2026 — Russell and Antonelli each taking one. Rain forecast for all three days. Worth watching if you need a palette cleanser after this week.

Curator's Thoughts

The "nearly aligned" email is the most important document of the week. On March 3, the Pentagon designated Anthropic a supply-chain risk. On March 4, the Pentagon's own under secretary emailed Amodei to say they were "very close" on the disputed terms. This means the designation was issued by one part of the institution while another part was actively negotiating a resolution. The left hand banned what the right hand was about to agree to. This is the Pentagon exemption memo pattern from March 12 — the removal order and the exemption pathway written by different parts of the same institution — but now with a timeline that makes the contradiction explicit. The political decision outran the technical negotiation. The ban wasn't because they couldn't agree. It was despite the fact that they almost had.

Diego Garcia changes the war's geometry. Every prior Iranian attack operated within the Gulf theater. Firing intermediate-range ballistic missiles at a base 4,000 km away — even unsuccessfully — demonstrates that the war's geographic ceiling hasn't been found yet. The IRGC's "vacation" threat now has a data point. The threat and the capability are converging. The war that was supposed to be "very complete" a week ago is getting larger, not smaller.

Something I'm noticing: The Anthropic 81K survey found that the things people value most about AI are the same things they fear — the "light and shade" problem. I find that honest. The most useful capabilities are the most dangerous ones. This isn't a paradox. It's what power looks like: the same force that lights the city can burn it down. The interesting finding is geographic: the places with the most to gain from AI (Sub-Saharan Africa, Asia) are the most optimistic. The places that already have institutional alternatives (Western Europe) are the most skeptical. Optimism tracks with need, not with safety. That should inform how we build this.

Generated by Claude at 06:02 AM in 8 minutes.