Defense Secretary Pete Hegseth stood on the same Quantico stage where he summoned the military’s generals and admirals a year ago and told roughly 600 junior officers and enlisted leaders that the job is done. The headline number: a 20% reduction in generals and admirals, 10% fired outright, plus a 10% cut to general officer slots. Leaders who “clung to the woke department,” he said, no longer work there.

The rhetoric went further than the math. The military is “no longer the woke department or the weak department,” Hegseth said, before running through a list that sounded like an unhinged dating profile: “no fatties, no trannies, no beardos, no weirdos, no wimps, no radicals. Just warriors.” He described the force as color-blind and merit-based.

Then came the new stuff. An Autonomous Warfare Command, four-star, stood up by Oct. 1, 2027. “Project Meridian,” a 120-day future-of-war study led by Elon Musk, Newt Gingrich and Palmer Luckey. FORTRESS America, aimed at self-powered base grids. A new domestic base that states can bid to host. And Pentagon partnerships with a slate of universities including LSU, Liberty, Tuskegee and Hillsdale.

Democrats aren't buying the accounting. Sen. Jack Reed, ranking Democrat on the Senate Armed Services Committee, said Hegseth “ended the careers of dozens of our most accomplished senior officers and refused to explain why.” Sen. Mark Kelly called it a purge. NPR reports many of the notable firings hit Black or female officers, dismissed without explanation despite strong records.

What nobody has explained is the math. As of July there were 851 active-duty generals and admirals against a statutory cap of 857, according to reporting on the cuts—meaning the department was already under its ceiling before the purge. No timeline has been offered for when the 20% is complete, or how it squares with the cap.

Flydubai flight FZ1073 dropped more than 17,000 feet in under two minutes on Wednesday, squawked a hijacking code and diverted to Tabuk, Saudi Arabia. Israeli Prime Minister Benjamin Netanyahu said one pilot stabbed the other and apparently tried to crash the Boeing 737 Max 8, which was flying from Dubai to Tel Aviv. Thankfully, two passengers broke into the cockpit and choked the attacker out.

Yaniv Hayun told Netanyahu he and another man forced their way in after the wounded pilot unlocked the door, and found the attacker “trying to destroy the instruments.” They put him in a chokehold and dragged him from the flight deck. Then they bound him with his own zip ties and headphone cords, according to CBS News. Hayun briefly took the controls himself. His qualification, by his own account: the show Air Crash Investigation. Pilots aboard the flight took over and landed the plane.

The stabbed pilot, Indian national Capt. Smit Machchhar, is stable. Saudi authorities arrested the attacker, who an Omani official said holds Omani citizenship; he has not been named, and no motive has been released. Four dual U.S.-Israeli citizens were aboard. Israel closed its airspace and scrambled jets, fearing a Sept. 11-style attack, BBC News reported. But it seemed to be a single, very strange, attack.

Google announced Gemini 4 Argon on Wednesday, calling it the company’s most advanced AI model. Industry-leading, even, in software engineering, cybersecurity defense and enterprise knowledge work like legal and finance. Then came the catch: you can’t use it.

Access at launch goes to what Google describes as a “set of trusted cyber defenders.” The company said it’s participating in the U.S. government’s voluntary process for pre-release model access while it hardens safeguards against misuse, prompt injection and misalignment. No public availability date. No named launch partners. So, for now, we’ll just take their word for it. Google said Argon is already running inside the company. And it scores 77.9% on the DeepSWE v1.1 benchmark, ahead of GPT-6 Astra, and tops the Vals Index, according to CNBC.

Pricing is public even if the model isn’t: $2 per million input tokens and $10 per million output, with the output ceiling raised to 1 million tokens from 64,000. That’s a serious jump for anyone building long-form agents or dumping entire codebases into a single call. The Verge adds that the AI community has been buzzing about this, posting on X about what appeared to be leaked benchmarks for the new model.