Formal verification has historically been limited to small codebases or required substantial expert effort; for example, the seL4 microkernel (~8,700 lines of C) took years of expert specification work before proofs mattered. AI tools are now being used to help write specs and proofs, but a proof is only as good as its model and specification.
Reuters reports that Romania is accelerating development of Black Sea capabilities following cargo ship attacks.
CNBC reports that Elon Musk has poured $366 million into midterm campaigns, with the Senate being his biggest political bet because control could affect investigations, spending, and regulations impacting SpaceX and Tesla.
CNBC reports Moderna shares fell after full data on its cancer vaccine, developed with Merck, leaked despite positive results.
CNBC, citing The Wall Street Journal, reports that Trump asked Musk to lift a Starlink slowdown; Musk had previously ordered the slowdown, and he is a critic of Ukrainian President Zelenskyy.
CNBC reports Meta has made clear Mark Zuckerberg will stay in California if the state's wealth tax passes, adding fuel to the debate ahead of the midterms.
EElon Musk has now put $366 million into these midterms, and CNBC says the Senate is his biggest bet.
SThat's not a bet. That's a down payment on who gets to investigate him.
EThat's exactly what the piece argues. Control of the Senate shapes investigations, federal spending, regulations on SpaceX and Tesla.
SSo he's not just buying ads. He's buying the committee that would subpoena him.
ELook, I get the cynicism. But a CEO spending to protect his companies from what he sees as political retaliation—that's not new.
SNo, it's just bigger. And the guy owns the platform where a lot of those ads run.
EFair point. But the other side has donors too. This is the system working as designed, for better or worse.
SThe system working as designed is exactly what worries me.
EOkay, we'll leave that one there. Different world entirely: Moderna's cancer vaccine data leaked, and the stock fell anyway.
SWait—the results were positive and shares still dropped?
ECNBC says the full data came out early, and the market didn't like what it saw in the details.
SSo the good news came with a catch.
EThat's biotech. A leak is also just a terrible look for a company that's supposed to control its data.
SAnd this is the vaccine they're developing with Merck, right? The one everyone's watching.
ERight. Positive results, but the stock reaction tells you investors wanted more than 'positive.'
SOr they hate the leak itself. If data can get out early, what else is loose?
EThat's a real concern. Speaking of things getting out: Trump reportedly asked Musk to lift the Starlink slowdown in Ukraine.
SWait, so Musk slowed it down, and the president had to ask him to turn it back on?
EThat's what the Wall Street Journal reported, and CNBC is running with it. Musk ordered the slowdown, then Trump asked him to reverse it.
SA private citizen controlling a war zone's internet, and the president has to negotiate with him.
EIt's not just internet. Starlink is how Ukraine coordinates a lot of its defense.
SSo one guy's mood can degrade a country's military operations. That's not power. That's a vulnerability.
EI'd argue it's both. And it's why the Senate race matters so much to him—he's got real leverage, and he knows it.
SWhich brings us right back to the $366 million. It's all connected.
EIt is. And here's one more piece: Meta says Zuckerberg is staying in California if the wealth tax passes.
SIs that a promise or a threat?
ECNBC frames it as adding fuel to the debate. The tax would hit billionaires hard, and Zuckerberg is making clear he won't leave.
SSo he's saying, 'Tax me, I'll stay.' That's actually... refreshing?
EOr he's daring California to try it, knowing other states would love to have him.
SBut that's the opposite of the usual billionaire playbook.
ETrue. Though staying in California with a wealth tax means writing a very large check every year.
SWhich he can afford. And it might actually help the tax's chances if voters see the rich won't flee.
EThat's the debate. Two weeks out, and it's only getting louder.
SMeanwhile, Anthropic just posted about using Claude to formally verify software at scale.
ENow that's interesting. Formal verification used to be this tiny, expert-only corner of computer science.
SRight—proving code has no certain bugs. Historically it only worked on small codebases, or took forever.
EThe seL4 microkernel is the classic example. About 8,700 lines of C, and it took years of expert work.
SSo if AI can write the specs and proofs faster, that's a genuine breakthrough.
EBut the proof is only as good as the model and the spec. Garbage in, garbage out.
SExactly. The proof only checks the code against the spec.
EStill, if this works at scale, it changes what 'safe software' even means.
SAnd it's a reminder that AI's biggest wins might be in the unglamorous stuff—verification, testing, proofs.
ENot chatbots. The boring stuff that keeps planes in the air.
SI'll take boring if it means fewer bugs.