Bell Test Demystified: How Experiments Proved Quantum Entanglement Is Real

Quick answer: A Bell test measures how strongly the results from two entangled particles agree when you measure each one far apart, using randomly chosen settings. Any theory built on local, pre-set properties ("hidden variables") caps the standard score, called S, at 2. Quantum mechanics allows up to 2√2, about 2.83. Experiments since 1972 have exceeded 2 again and again. In 2015, three teams closed the major loopholes, and the work earned the 2022 Nobel Prize in Physics. In 2025 and 2026, the Bell test became a commercial tool: it now certifies public random numbers (NIST's CURBy), secures encryption keys over 100 km of fiber (USTC), and verifies quantum hardware.

New to entanglement? Start with our primer, Quantum Entanglement Explained. This Architect piece goes one layer down: how the experiment works, how physicists closed each escape route, and why the test itself is becoming a product.

A Bell test is a statistics game played with entangled particles that no pre-arranged strategy can win as often as quantum physics does.

The setup has three parts:

Here is the intuition. Imagine Alice and Bob are contestants who can agree on any strategy before they're separated, but can't communicate once questions start. Each gets a random question and must answer yes or no. Bell showed that for the right set of questions, the best possible pre-agreed strategy wins at most 75% of the time. Entangled particles win about 85% of the time. No amount of clever planning closes that gap. Only a shared quantum state does.

Section summary: A Bell test compares measurement results from distant, entangled particles under random settings. If they agree more often than any pre-set strategy allows, the particles were not carrying hidden instructions.

The Bell test exists because Einstein bet that quantum mechanics was incomplete, and John Bell found a way to check.

In 1935, Albert Einstein, Boris Podolsky, and Nathan Rosen (EPR) argued that if quantum mechanics were complete, measuring one particle would instantly fix the state of a distant partner. They concluded the theory must be missing "elements of reality": pre-set properties we simply couldn't see. Einstein later called the alternative "spukhafte Fernwirkung" ("spooky action at a distance") in a 1947 letter to Max Born. That phrase is often wrongly credited to the EPR paper itself.

For three decades the debate stayed philosophical. Then in 1964 John Stewart Bell, a physicist at CERN, proved that local hidden-variable theories and quantum mechanics make different, measurable predictions. In 1969, Clauser, Horne, Shimony, and Holt turned Bell's idea into a lab-ready formula, the CHSH inequality, which remains the workhorse today.

The CHSH test combines four correlation measurements, from two settings at each station, into a single score, S:

If an experiment reliably measures S above 2, it rules out local hidden variables. Every result below should be read against that line.

Section summary: Bell's 1964 theorem and the 1969 CHSH formula turned the EPR debate into a single testable number. Classical physics caps S at 2. Quantum mechanics allows about 2.83.

The proof came in stages over 50 years. Each experiment closed one way a classical explanation could still slip through.

Aspect's 1982 innovation was speed. Earlier tests fixed the polarizer angles before the photons left the source, so a slower-than-light signal could, in principle, coordinate the results. Aspect's team used acousto-optical switches that changed the analyzer direction while the photons were in flight. That made it much harder for one station to "know" the other's setting in time. The violation held.

In 2015, three independent groups on two continents closed the locality and detection loopholes in the same experiment. Delft used electron spins in diamond, separated by 1.3 km, and measured S = 2.42 over 245 trials (Hensen et al., Nature). NIST and Vienna used photons with superconducting detectors efficient enough to detect most particles. Three different platforms reached the same answer, making the result hard to dispute.

Section summary: From Freedman and Clauser (1972) to Aspect (1982) to the 2015 loophole-free tests and the 2022 Nobel, every major Bell experiment has violated the classical limit. The question of whether entanglement is real is closed.

A loophole is not a flaw in quantum mechanics. It is a gap in an experiment's design that would let a classical explanation survive. There are three that matter.

If one station could signal its setting or result to the other before the measurement ends, even at light speed, classical coordination is possible. The fix is spacetime separation: choose settings and finish measurements faster than light can travel between stations. Aspect started this, Weihs and Zeilinger tightened it at 400 m, and Delft's 1.3 km separation sealed it. ETH Zurich's 2023 experiment completed each trial in about 107 nanoseconds across a 30 m cryogenic link, within the roughly 110 ns light-travel budget.

If detectors miss too many particles, the ones they catch could be an unrepresentative subset that only looks quantum. The fix is efficiency. Photon experiments need very high overall detection efficiency, which superconducting nanowire and transition-edge detectors made possible. NIST and Vienna crossed that bar in 2015. Ion, atom, and superconducting-qubit experiments avoid the problem because they read out nearly every trial.

If some hidden cause influenced both the particles and the "random" setting choices, the test would be rigged from the start. This is also called the measurement-independence or "superdeterminism" loophole. It can never be fully closed, only pushed back in time.

That is what the Cosmic Bell experiments did. They did not time measurements using astronomical signals. Instead, two telescopes on La Palma in the Canary Islands read the color of light from two distant quasars, and that color chose each station's measurement setting. The light had left its sources 7.8 and 12.2 billion years ago. Any conspiracy would have to have been arranged before that light was emitted. The 2018 results (Physical Review Letters) still violated Bell's limit, with odds of a classical explanation of about 10⁻²⁰ for the stronger run.

"As strange as quantum mechanics may seem, it continues to match every experimental test we can devise," said David Kaiser, professor of physics and the history of science at MIT, on the 2018 result.

A related 2016 effort, the BIG Bell Test, used choices typed by about 100,000 human volunteers worldwide as the random source.

Section summary: Locality was closed with distance and speed, detection with high-efficiency detectors, and freedom of choice was pushed back billions of years using quasar light. No loophole remains that a mainstream physicist treats as a live explanation.

The commercial value of a Bell test is this: a violation proves something about the outputs even if you don't trust the device that made them.

A normal random number generator or encryption box asks you to trust the vendor's engineering. A Bell-based system flips that. If the logged results violate the CHSH inequality, then by physics the outputs:

This is called device-independent (DI) security. You check the statistics, not the box. For regulated industries, sovereign buyers, and anyone worried about supply-chain tampering, that is a new kind of assurance.

Three markets are doing this now.

CURBy, the Colorado University Randomness Beacon, is a public service that delivers random numbers certified by a loophole-free Bell test. NIST and the University of Colorado Boulder built it and published it in Nature in June 2025 (Kavuri et al.). Entangled photons travel by fiber to labs at opposite ends of a hall, and the Bell test runs 250,000 times per second. In its first 40 days, CURBy produced certified output on 7,434 of 7,454 attempts (99.7%), with 512 random bits per run. A blockchain-style hash chain, the open-source Twine protocol, lets anyone audit each output.

"CURBy is one of the first publicly available services that operates with a provable quantum advantage," said Krister Shalm, physicist at NIST.

Target uses include jury selection, audit sampling and lotteries: anywhere a regulator needs randomness that no insider could have rigged.

A related approach: in March 2025, Quantinuum, JPMorganChase, Argonne and Oak Ridge National Laboratories and UT Austin reported certified randomness from a trapped-ion quantum computer (Nature). That method uses random circuit sampling, not a Bell test, but it targets the same buyer: financial institutions that need provable randomness.

DI-QKD uses a Bell violation to prove an encryption key is private, even if the hardware came from an untrusted supplier. Until recently, it worked only across a lab. In 2022, Oxford extracted a key over about 2 meters using trapped ions, and LMU Munich reached about 700 meters with atoms.

In early 2026, a USTC team led by Pan Jianwei, Zhang Qiang and Bao Xiaohui reported DI-QKD over 100 km of fiber in Science, using single rubidium atoms with frequency conversion to telecom wavelengths. They measured S ≈ 2.61 at 11 km and extracted about 135 kilobits of finite-key secure material over a 624-hour run.

Two caveats matter for investors:

"[The rates are] abysmally small – producing less than one bit of secure key every 10 seconds," said Steve Rolston, physicist at the University of Maryland, in coverage reported by PostQuantum.com.

Bell tests are becoming a benchmark for whether quantum hardware produces real entanglement, not just correlated noise. In May 2023, Andreas Wallraff's group at ETH Zurich ran the first loophole-free Bell test with superconducting qubits, the same technology IBM and Google use. They performed it across a 30 m cryogenic link cooled below 50 millikelvin. Over more than 1 million trials, they measured S = 2.0747, with a P value below 10⁻¹⁰⁸ (Nature).

The S value is modest, but the implication is large. It shows Bell-grade entanglement between separate refrigerators, a building block for modular, networked quantum computers. The authors point to device certification, key distribution, and randomness generation as next uses.

Section summary: Device independence turns a Bell violation into an auditable guarantee. Certified randomness is live, DI-QKD reached metro distance in 2026 at very low rates, and Bell tests now verify entanglement on quantum-computer hardware.

Capital is moving from proving entanglement to distributing and certifying it. The McKinsey Quantum Technology Monitor 2026 reports $12.6 billion invested in quantum startups in 2025, up 6.3x from 2024. Private capital made up 97%, and the top ten deals accounted for about 60%. Analysts caution that Chinese startup activity is undercounted.

What this means for allocators: the Bell test itself is not a business. The entanglement infrastructure that makes Bell-grade guarantees deliverable is. That means networking hardware, quantum memories, orchestration software and certification services. Those layers sell to every qubit modality, much like the noise-management layer covered in our previous Architect piece.

Section summary: 2025 to 2026 funding is concentrating on entanglement distribution (Nu Quantum, Qunnect, Aliro) and on consolidation (IonQ). Certification services are an emerging, still largely public-sector category.

"Unhackable" marketing. Bell tests certify the physics, not the full system. Software, authentication, and side channels can still fail. Treat absolute security claims as a diligence warning.

QKD is not PQC. The NSA still does not recommend QKD for national security systems, and its 2026 guidance rewrite kept that verdict. The UK NCSC (August 2025) will not support QKD for government or military use. France, Germany, the Netherlands and Sweden jointly advise migrating to post-quantum cryptography first. Even DI-QKD needs classical authentication.

Lab vs. field. The 100 km DI-QKD result used coiled fiber in one lab. Ask whether a claim was shown across separated, deployed sites.

S-value inflation. A "violation" with weak statistics, or with loopholes left open, isn't certification. Ask which loopholes were closed and what the P value was.

Rate vs. reach. Distance headlines often hide key rates that are orders of magnitude below what commercial traffic needs.

Concentration risk. With roughly 60% of 2025 capital in ten deals, valuations in the networking layer may be priced off a few comparables.

Separate the proof from the product. Entanglement is proven. The question is whether a company sells a certified output (randomness, keys, verification) that a buyer will pay for.

Run PQC migration now. Treat DI-QKD as a 5+ year option. Don't let quantum-security pitches delay post-quantum cryptography upgrades.

Pilot certified randomness where audits matter. Gaming, lotteries, jury and audit selection, and financial sampling can use public beacons like CURBy today at low cost.

Back the picks-and-shovels of entanglement. Distribution hardware, quantum memory and network orchestration serve computing, sensing and security together.

Use Bell metrics in hardware diligence. For networked or modular quantum systems, ask for a CHSH value between modules and which loopholes were closed.

Expect three developments. DI-QKD will move from coiled fiber to deployed links between separate sites, finally closing locality in the field. Certified randomness will become a paid service, with Quantinuum signaling commercial integration and public beacons multiplying. And networked quantum computers will cite Bell violations between modules as proof that interconnects work. Underwrite teams that publish key rates and loophole status, not just distances.

The Bell test forces a choice: either the universe isn't locally pre-scripted, or something stranger is going on. A violation means nature cannot be described by particles carrying hidden, local instructions. It does not mean information travels faster than light. The correlations only appear when the two logs are compared over an ordinary channel.

The everyday analogy of two coins that always land opposite each other is useful but incomplete. Classical coins can be correlated through a shared past. A Bell violation shows that entangled particles are correlated more strongly than any shared past allows. That gap, between 2 and 2.83, is where the physics, and now the business, lives.

The Bell test started as a challenge to quantum mechanics. Fifty years later, it is quantum mechanics' audit trail. For decision-makers, the lesson is this: the strangest property in physics is becoming the most verifiable one.

A Bell test is an experiment that measures correlations between entangled particles under randomly chosen settings at separated locations. If the correlations exceed the classical limit (S > 2 in the CHSH version), no theory based on local, pre-set properties can explain them.

John Clauser (with Stuart Freedman, 1972), Alain Aspect (1982), and Anton Zeilinger (1998 onward) led the key experiments and shared the 2022 Nobel Prize in Physics. Loophole-free tests by Delft, NIST, and the University of Vienna in 2015 removed the remaining major objections.

Bell's inequality, derived by John Stewart Bell in 1964, is a mathematical limit on how strongly the results from two separated systems can correlate if local hidden variables set those results. The CHSH form caps the score at 2. Quantum mechanics predicts up to about 2.83.

There are three main ones. The locality loophole (stations could communicate), the detection loophole (missed particles could bias results), and the freedom-of-choice loophole (settings might not be truly independent). The first two were closed together in 2015. The third was pushed back billions of years by the 2018 Cosmic Bell test using quasar light.

No. A Bell violation shows correlations stronger than any local hidden-variable theory allows, but neither party can use them to send a message. The correlation only appears when results are compared over a normal channel.

Bell tests enable device-independent security. NIST's CURBy beacon publishes Bell-certified random numbers. Device-independent QKD reached 100 km in 2026 (USTC). Bell tests also verify entanglement between quantum-computer modules (ETH Zurich, 2023).

Not yet. The 2026 USTC demonstration produced about 0.06 secure bits per second at 11 km on coiled lab fiber. Enterprises should prioritize post-quantum cryptography and treat DI-QKD as a long-term option for high-assurance links.

Get weekly insights on emerging tech trends, funding rounds, and strategic opportunities, straight from the field. As a Fractional CTO working across AI, blockchain, quantum, IoT, and Web3, I share what I'm seeing in the trenches, not what I'm reading in press releases.

Subscribe to Unbound →

Barbara Bickham is Founder & CTO of Trailyn Ventures and writes Unbound, weekly deep-tech intelligence for executives, investors, family offices, and LPs.

NobelPrize.org, "The Nobel Prize in Physics 2022 — Popular information": https://www.nobelprize.org/prizes/physics/2022/popular-information/

Physics Today, "Three groups close the loopholes in tests of Bell's theorem": https://physicstoday.aip.org/news/three-groups-close-the-loopholes-in-tests-of-bells-theorem

Storz et al., "Loophole-free Bell inequality violation with superconducting circuits," Nature 617, 265–270 (May 10, 2023): https://pmc.ncbi.nlm.nih.gov/articles/PMC10172133

ScienceDaily / MIT, "Light from ancient quasars helps confirm quantum entanglement" (Aug 20, 2018): https://www.sciencedaily.com/releases/2018/08/180820114121.htm

Rauch et al., "Cosmic Bell Test Using Random Measurement Settings from High-Redshift Quasars" (arXiv:1808.05966): https://arxiv.org/abs/1808.05966

NIST, "NIST and Partners Use Quantum Mechanics to Make a Factory for Random Numbers" (Jun 11, 2025); Kavuri et al., Nature, DOI 10.1038/s41586-025-09054-3: https://www.nist.gov/node/1883571

The Quantum Insider, "NIST-led Team Uses Quantum Mechanics to Make a Factory for Random Numbers" (Jun 11, 2025): https://thequantuminsider.com/2025/06/11/nist-led-team-uses-quantum-mechanics-to-make-a-factory-for-random-numbers/

Quantinuum, "Quantinuum Introduces First Commercial Application for Quantum Computers" (Mar 26, 2025): https://www.quantinuum.com/blog/quantinuum-introduces-first-commercial-application-for-quantum-computers

Lu et al., "Device-independent quantum key distribution over 100 km with single atoms" (arXiv:2602.09596): https://www.alphaxiv.org/abs/2602.09596

Chinese Academy of Sciences, "Chinese Researchers Achieve Device-Independent Quantum Key Distribution over 100 km": https://english.cas.cn/special-reports/2026q1/202604/t20260427_1158169.shtml

PostQuantum.com, "China pushes DI-QKD to 100 km" (analysis and expert comment): https://postquantum.com/security-pqc/china-di-qkd-100/

PostQuantum.com, "NSA QKD guidance rewrite" (Oct 2026): https://postquantum.com/quantum-policy/nsa-qkd-guidance-rewrite/

PostQuantum.com, "McKinsey Quantum Technology Monitor 2026": https://postquantum.com/industry-news/mckinsey-quantum-monitor-2026/

I-Connect007, "Nu Quantum Raises $60M Series A" (Dec 15, 2025): https://iconnect007.com/article/148091/nu-quantum-raises-60m-series-a-in-largest-financing-round-for-quantum-computer-networking/148088/aep

Quantum Computing Report, "Aliro Raises $15 Million to Advance Physics-Based Network Security" (Feb 18, 2026): https://quantumcomputingreport.com/aliro-raises-15-million-to-advance-physics-based-network-security/

Channel Insider, Qunnect $10M Series A extension (Jun 25, 2025): https://www.channelinsider.com/news-and-trends/quantum-network-regulated-industries-funding

Converge Digest, "IonQ to Acquire Lightsynq" (May 2025): https://convergedigest.com/ionq-to-acquire-lightsynq-accelerating-quantum-interconnects/