The Big Picture
Iran is running a naval war inside a ceasefire America agreed to extend — seizing ships, firing on vessels, and mining the Strait of Hormuz — while the White House calls it a non-event because "they weren't U.S. ships." Meanwhile, China is rehearsing the invasion of Taiwan with live missiles, North Korea's hackers just stole over half a billion dollars in a single month to fund Kim's weapons programs, and a Chinese AI model that fits on a laptop is closing the gap on American frontier technology. Our adversaries are not resting. The question is whether Washington is paying attention.
Today's Stories
IRAN IS RUNNING A BLOCKADE INSIDE OUR CEASEFIRE — AND WASHINGTON IS LETTING IT HAPPEN
Iran's Revolutionary Guard seized two commercial vessels — the MSC Francesca and the Epaminondas — in the Strait of Hormuz on Wednesday, and fired on a third. This happened hours after President Trump extended the ceasefire indefinitely. The White House response? Not a violation, because they "weren't U.S. ships." That doctrinal gap is the ballgame. Tehran is conducting active maritime warfare under a ceasefire umbrella Washington refuses to pierce. The Pentagon has told Congress that 20 or more mines may have been laid in the strait, and full clearance could take six months after any peace deal. Brent crude broke $100 intraday. WTI topped $94. Roughly one-fifth of global oil transits Hormuz — that's your gas prices, your freight costs, your grocery bills. Fifteen American service members have died in this conflict. Iran still has roughly half its ballistic missile stockpile intact, per the Defense Intelligence Agency. And the Navy Secretary just resigned in the middle of it all. This is not a managed situation. This is a slow-motion crisis being treated like a scheduling problem.
CHINA FIRED SEVEN MISSILES AROUND TAIWAN. ADMIRAL PAPARO CALLED IT A DRESS REHEARSAL.
The People's Liberation Army fired seven missiles into declared no-fly zones around Taiwan during an exercise it branded "Justice Mission 2025," with reporting indicating 27 total missiles fired in or around Taiwan's contiguous zone. No injuries — this time. But Admiral Samuel Paparo told Congress plainly: "Beijing's aggressive maneuvers around Taiwan are not just exercises — they are dress rehearsals for forced unification." The operational detail that matters is what these drills are actually testing: kill chains, sensors, datalinks, electronic warfare, and timed cross-domain effects. Every drill forces U.S. and Taiwanese radar and satellite systems to light up, revealing collection patterns to Chinese intelligence. Each rehearsal is more sophisticated than the last. China is not bluffing. It is training. And while Washington is consumed with Hormuz, Beijing is running the clock on Taiwan.
NORTH KOREA STOLE $575 MILLION IN ONE MONTH. IT'S FUNDING NUCLEAR WEAPONS.
Kim Jong Un's Lazarus Group has now stolen over $575 million in cryptocurrency in April alone — $285 million from Drift Protocol on April 1st, and $290 million from KelpDAO last weekend. The KelpDAO attack didn't break the smart contract. It poisoned the infrastructure that verifies whether transactions are real, compromising RPC nodes and using a DDoS attack to force failover onto compromised servers. The UN panel that tracked North Korean sanctions evasion — before being disbanded — estimated illicit cyber operations fund roughly 40 percent of Pyongyang's weapons programs. Half a billion dollars in a month is not a rounding error on a nuclear program. This is state-sponsored financial warfare, and American crypto platforms are the soft target. The playbook is now public. Expect a third nine-figure heist before May if the industry doesn't respond.
CHINA'S AI JUST SHRANK TO LAPTOP SIZE — AND IT'S CLOSING THE GAP
Alibaba released Qwen3.6-27B, an open-source AI model that, according to Alibaba's own benchmarks, outperforms its previous 397-billion-parameter flagship on key coding and reasoning tests. The old model weighed 807 gigabytes. The new one: 55.6 gigabytes — small enough to run on a high-end consumer GPU. At compressed settings, it fits in 16.8 gigabytes of VRAM. This is the AI competition in concrete form. American frontier labs are building models that require server farms; China is building models that run on developer laptops, distributing them freely under open licenses, and closing the performance gap. Simultaneously, Alibaba's Tencent is in talks to fund DeepSeek — China's efficiency-first AI lab — at a $20 billion valuation, giving it the capital to buy chips aggressively. The technology competition with China is not a future problem. It is today's problem, and the scoreboard is moving.
THE MOST DANGEROUS AI MODEL EVER BUILT LEAKED ON DAY ONE
Anthropic built an AI model called Claude Mythos — described internally as the most offensively capable model it has ever created — and restricted it to a small pool of vetted companies for cybersecurity research. It leaked through a contractor environment on approximately the same day it launched. According to UK AI Security Institute evaluations, Mythos succeeds on expert-level cyberattack simulations 73 percent of the time. It is the first model to complete a 32-step corporate network attack end-to-end. Anthropic used it to identify thousands of zero-day software vulnerabilities — over 99 percent of which remain unpatched. Unauthorized users accessed it through a third-party vendor by guessing a URL, with credentials reportedly linked to a separate breach of an AI recruiting startup. The companies with authorized access include Apple, Amazon, Cisco, Google, Microsoft, and JPMorgan. If any of those findings have been exfiltrated, the exposure is systemic. This is what happens when the most powerful offensive cyber tools ever built are managed with contractor-grade access controls.
What to Watch
- [CONFIRMED] If Iran formally rejects peace talks in writing within the White House's reported 3-to-5-day window, Brent crude could test $110 intraday and the Lloyd's war-risk market may trigger a durable repricing of Gulf transit insurance — locking elevated energy costs into the global economy for months. (Confirmed: based on reported Pentagon briefing timeline and Lloyd's market mechanics)
- [ASSESSED] If the Lazarus Group's RPC-poisoning attack method is adopted as a template — which it now can be, since the playbook is public — a third nine-figure cryptocurrency heist before May is probable, with DeFi platforms running single-verifier configurations as the most exposed targets. (Assessed: analytical projection based on confirmed attack methodology)
- [ASSESSED] If China's Qwen3.6-27B benchmarks are independently verified by third-party researchers on SWE-bench's leaderboard, the economics of American AI dominance shift materially — on-premise Chinese AI agents become viable for any engineering team worldwide, eroding the frontier lab moat that U.S. companies currently hold. (Assessed: analytical projection pending independent benchmark confirmation)
- [SPECULATIVE] If the unauthorized access to Anthropic's Claude Mythos Preview involved exfiltration beyond Discord curiosity, the 40-plus organizations in the authorized program — including Apple, Microsoft, and Cisco — face immediate security review pressure, and every AI lab's contractor access policy becomes a board-level national security question. (Speculative: Anthropic investigation ongoing, scope unconfirmed)
The Closer
Iran is seizing ships under our ceasefire, China is rehearsing Taiwan's invasion with live missiles, and North Korea just stole enough money to run a serious weapons program — all in the same 24 hours. Our adversaries have a strategy. The question Americans need to be asking their elected officials is simple: do we?