Did Google just prove Entanglement exists mathematically — by using Entanglement physically between qubits!

A five-year-old's guide to the game no classical computer can win — the shared cheat sheet that cannot exist — and how Google's Willow won it 98.3% of the time, turning a scoreboard into a proof of entanglement.
Did Google just prove Entanglement exists mathematically — by using Entanglement physically between qubits!

There is a children’s game that no one in the universe should be able to win. Not you, not me, not the biggest supercomputer on Earth. The rules of arithmetic itself say so.

This week, on the cover of Science, Google’s quantum chip Willow won it — 900,000 times in a row, scoring 98.3%. And the way it won is the whole story: the game can only be won if entanglement is real. So the score is not just a score. It is a proof.

Let me explain the game the way you’d explain it to a five-year-old, because honestly, that’s the only way any of us really understand it.

The game

Picture a tic-tac-toe board — three rows, three columns, nine boxes. Every box holds one of two answers: +1 or −1. Think of them as “yes” and “no,” or heads and tails. Two answers, nothing else.

Two players, Alice and Bob, play as a team. But the moment the game starts, they are put in separate rooms. No talking, no texting, no peeking.

Each round, the referee asks Alice about one row and Bob about one column. Alice must announce her row’s three answers, and they must multiply together to +1. Bob must announce his column’s three answers, and they must multiply to −1. And one more thing: the box where Alice’s row and Bob’s column cross belongs to both of them — their answers for that box must match.

Get the products right and the crossing box matching, and the team wins the round.

Why you cannot win this game

Before the game, Alice and Bob are allowed to plan. The obvious plan: fill in the whole board in advance, memorize it, and just read out your row or column when asked. A cheat sheet.

Here is the trap, and a five-year-old can follow it. If every row multiplies to +1, then the whole board, multiplied together, comes to +1. But if every column multiplies to −1, the whole board comes to −1. The whole board cannot be both +1 and −1. So no completed board exists that obeys all the rules. Every cheat sheet ever written has at least one broken row or column hiding in it.

But wait, you say — don’t Alice and Bob have the same sheet? They do. And that is precisely the point: the problem was never coordination. The problem is the sheet itself.

Look at the board in the picture. That is the best cheat sheet anyone in the universe could ever write — every row multiplies to +1, columns one and three multiply to −1. Now read column two out loud: +, −, −. It multiplies to +1, and the rule demands −1. So whenever the referee asks Bob for column two, faithful Bob, reading the shared sheet exactly as agreed, loses on the spot — not because he and Alice failed to coordinate, but because the sheet they share is broken, and every sheet is broken somewhere.

Fine, you say — let Bob patch column two. Flip the bottom entry, and his column now multiplies to −1. Fixed? Now Bob’s sheet no longer matches Alice’s. When the referee happens to ask Alice for row three in that same round, the crossing box says −1 on her sheet and +1 on his. Mismatch. Lose that round instead. Every patch just moves the loss from one round to another. It never removes it. The sheet is not badly written — a perfect sheet is arithmetically impossible, and identical copies of an impossible sheet are still impossible.

That means the best any team with a fixed plan — any person, any classical computer, any strategy written down in advance — can score is 8 rounds out of 9. Eighty-eight point nine percent. That is not an engineering limit. That is arithmetic. It is a ceiling built into the numbers themselves.

What Google did

Google’s team didn’t give Alice and Bob a cheat sheet. They gave them something else: two pairs of entangled qubits on the Willow processor — four qubits in total. Two “belong” to Alice, two to Bob.

Entanglement means this: the qubits do not carry answers of their own. There is no little +1 or −1 written inside them waiting to be read out. Each answer is born at the moment the question is asked — and which answer is born depends on which question, row or column, came with it. Nothing is fixed in advance, so the arithmetic trap — which only catches things that are fixed in advance — has nothing to catch.

Then they played. All nine combinations of row and column, 100,000 times each. 900,000 rounds.

Willow won 98.3% of them.

Not 100% — real hardware hums and jitters, and noise costs a round here and there. But 98.3% against a mathematical ceiling of 88.9% is not a small beat. It is a beat that classical physics says cannot happen at all.

Now read the title again

Did Google just prove entanglement exists mathematically — by using entanglement physically between qubits?

That is exactly the beautiful loop here. The mathematics of the game draws a line and says: nothing that works by fixed, pre-written answers can cross this line. The only way across is if the players’ answers genuinely do not exist until the moment of the question, yet still arrive in agreement — which is the definition of entanglement. So Google built the physical thing, entangled real qubits, played the game 900,000 times, and the scoreboard did the proving. The math defined the test. The physics passed it. The pass is the proof.

One honest correction, because it matters: Alice and Bob do not “know” what the other is doing. Bob, looking only at his own answers, sees pure randomness — he cannot tell which row Alice was asked, and no message of any kind travels between the rooms. This trick can never be used to send information. The magic only appears afterwards, when the referee lays the two answer sheets side by side and finds them agreeing far beyond what any pre-written reality allows. Agreement without a script, and without communication. That is the whole wonder.

Why this is bigger than a game

Entanglement has been demonstrated before, in delicate little tabletop experiments. What is new here is scale and stubbornness. Google ran companion tests of the same underlying property — physicists call it contextuality, the idea that a measurement’s answer depends on the company it keeps — across bigger and bigger systems, all the way up to tasks using every one of Willow’s 105 qubits, and the quantum behaviour refused to wash out. On one standard score for it, where any fixed-answer world caps at 4 and quantum theory allows at most 6, Willow scored 5.59.

This matters for a practical reason, not just a poetic one. A growing body of theory says contextuality is not a curiosity — it is the fuel. It may be the actual ingredient that lets quantum computers do things classical computers can’t. Google has now shown the fuel is still in the tank at 105 qubits, and they’ve handed engineers a sensitive new ruler: a game so unforgiving that tiny hidden errors in the hardware show up in the score.

The fine print

Keep three caveats, because five-year-olds deserve honesty too. This is one specially designed game, not a general victory over classical computers — nobody has replaced your laptop. The 98.3% is not the ideal 100%; noise is still the enemy, and taming it is the field’s real long war. And while the work is peer-reviewed and on the cover of Science, it is Google’s own team reporting Google’s own chip — independent replication is still to come.

But none of that dims the central fact. A game was designed whose rules make victory mathematically impossible for anything that plays by fixed answers. A machine just won it, 900,000 rounds deep, by being entangled — really, physically entangled.

The board is only three by three. The proof is somewhat larger.

Sources: Google, “How we’re putting one of quantum mechanics’ deepest features to the test at scale,” blog.google, Nov 19, 2026 — https://blog.google/innovation-and-ai/models-and-research/quantum-computing/putting-one-of-quantum-mechanics-deepest-features-to-test-at-scale/ · “Quantum-classical separation in bounded-resource tasks arising from measurement contextuality,” Science, Nov 19, 2026, DOI 10.1126/science.aec3042


Write a comment