AI Daily Grind // AI news, freshly ground daily

← All lessons

🛡️ Security · ⏱️ 2 min Cappuccino

What is Formal verification?

Illustration: Formal verification
Simply put, like I'm 5 years oldTesting code is like tasting a soup a few times to see if it is good. Formal verification is like checking every step of the recipe so carefully that you can be sure it is right, without tasting.

📖 The slightly longer version

Normally, people check software by testing it: they try many inputs and see if anything breaks. But tests only cover the cases someone thought of. Formal verification is different. You write down, in exact math, what the code must do. Then you build a proof, a chain of logical steps, that the code always does it, for every possible input.

A human can make mistakes in a proof, so a special program called a proof checker reads every step and refuses anything that does not follow. Lean is one such language: you write the proof in it, and the computer checks the math. Writing these proofs is slow and hard work for people. That is why it is big news when AI agents, small AI programs that work on a task by themselves, write them.

How formal verification works

  1. Write the rulesPeople describe in exact math what the code must always do.
  2. Build a proofA human or an AI writes logical steps showing the code follows the rules.
  3. Computer checksA proof checker like Lean tests every single step and rejects mistakes.
  4. Proof acceptedThe code is proven to match the rules for every possible input.
Why it matters

Google DeepMind says its AlphaProof Code agents wrote three million lines of Lean proofs to verify Signal's core encryption, the math that keeps your messages private. If AI can make this kind of checking cheap, the apps you trust with secrets could become much safer. This is the good side of AI and code, the opposite of fast agents shipping apps nobody checks.

⚠️ What to watch out for

A proof only shows the code matches the written description, called the specification. If the description is wrong or misses something, the proof can be perfect and the software still unsafe. Also, three million lines is far too much for humans to read, so we trust the proof checker and the specification, and we only have Google's own blog post on the details so far.

📰 Where it showed up

This notion came up in A trillion-dollar IPO, a new safety law, and 35,000 open doors. More lessons: Learn.

🎯 Keep learning

New 🛡️ Security · 2 min Latte What is API key? An API key is like the key to a shared kitchen where every meal you cook goes on your bill. If someone copies your key, they can cook all night, and you pay. Learn it → New 🛡️ Security · 2 min Latte What is Penetration Testing? It is like asking a friend to try every door and window of your house, so you find the one that does not lock before a burglar does. Learn it → 🛡️ Security · 2 min Latte What is Privilege Escalation? Imagine a hotel guest whose room key, because of a small mistake, also opens the manager's office and every other room. Privilege escalation is when a computer user finds a mistake like that and gets keys they were never supposed to have. Learn it →

Sources

Written on 2026-11-11. This explanation was accurate on that day. AI changes fast, so check the sources and the date before you rely on it.

Get The AI Daily Grind by email, every morning, in simple words.