Twenty-five years after the September 11 attacks, no one has been convicted or tried for them. Khalid Sheikh Mohammed, the alleged mastermind, and four alleged al-Qaida co-defendants are still at Guantánamo Bay, still in pretrial hearings. Victims’ families told reporters that the delay is “unconscionable.”

The case is on its fifth military judge. Lt. Col. Michael Schrama set a June 2028 trial date last month, then days later ruled Mohammed’s 2007 FBI confession inadmissible because it was tainted by torture. The Justice Department has said Mohammed was waterboarded 183 times at CIA black sites—a fact that has poisoned the evidentiary well since the day the commissions convened.

The list of collapses is long. A 2011 plan to try the men in a Manhattan federal court was scrapped. A 2021 trial date came and went. Ramzi bin al-Shibh was severed from the case in 2023 after being found mentally unfit. A 2024 plea deal that would have taken the death penalty off the table was reversed by the Biden administration, and the litigation over that reversal is still grinding. And what’s unresolved matters as much as what’s happened. It isn’t clear whether the June 2028 date survives the confession ruling, where the death penalty question now stands, or how the government intends to proceed on the plea deal fight.

The evidence speaks for itself. Critics of civilian trials in the case warned that jurors could not bear to relive the horror of Sept. 11 or that classified evidence could not be handled safely. That was wrong. Federal courts had already convicted dozens of terrorists (including Ramzi Yousef, mastermind of the first World Trade Center bombing, and Zacarias Moussaoui, once called the 20th hijacker) using lawful, tested procedures designed to balance the public’s right to see powerful evidence, the defendant’s right to a fair trial and the nation’s interest in protecting intelligence-gathering sources and methods.

The anniversary arrives as the TSA rolls back post-9/11 rules on gate access and liquids, citing better screening technology. The security state built after that morning is quietly relaxing … but the trial it produced hasn’t even started.

President Donald Trump stood up at the Republican National Committee’s first-ever midterm convention in Dallas and promised every adult U.S. citizen a $5,000 “dividend”—on the condition that the GOP hold both the House and the Senate in November. No funding source, no bill. He told the media on Thursday he doesn’t think Congress would even need to approve it, then repeated the pledge the following night.

Run the math: Roughly 245 million adults at $5,000 apiece puts the tab past $1 trillion. Tariff revenue, the president’s usual answer, doesn't come close. Tax Foundation economist Erica York told CBS News tariffs bring in about $125 billion a year (or a tenth of what’s needed). Borrowing the rest would push the deficit toward $3 trillion, up from about $1.8 trillion. All of it landing while August inflation was expected to print at 3.3%, which is the opposite of the moment you hand out cash.

The criticism came from both directions. Republicans Thomas Massie, Susan Collins and Chip Roy pushed back; Democrats dismissed it as an empty promise. On the law, it’s murkier than either side suggests. Election law expert Rick Hasen told PBS NewsHour that Trump can’t do this alone but that the promise itself is likely protected campaign speech. The Guardian reports whether the dividend would violate federal law remains open, so the pledge survives on a technicality: promising is legal, paying is the hard part.

An AI researcher quit Anthropic this week and posted his reasoning on X: “the people building AI earnestly believe that it could kill us all by the end of the decade.” And one of Anthropic’s senior safety executives publicly agreed with him.

As WIRED's Uncanny Valley podcast points out, the doom pitch isn’t new—Anthropic has built its whole identity on the idea that this technology is dangerous. Which, admittedly, is a strange thing to say while shipping it. What’s new is the timing. The resignation landed while AI agents are behaving unpredictably and OpenAI is claiming its model cracked a Clay mathematics problem, a claim some academics dispute. Meanwhile the labs are racing toward recursive self-improvement: AI that makes better AI.

As Will Knight says on the pod, “I’ve always felt that a lot of the people who are worrying about AI killing everybody tend to conflate whether something's possible for whether it’s likely, and then they throw a percentage that is kind of pulled out of thin air.”