ETH Zurich Creates First Certifiably Perfect Random Numbers

2. Creating entanglement across a 30-meter link. Two superconducting quantum chips are cooled to near absolute zero and connected by a 30-meter cryogenic link . They are placed into an entangled state, meaning measurements on one chip instantly correlate with the state of the other—a hallmark of quantum non-locality.

3. Certifying randomness with a loophole-free Bell test. The weak randomness determines the measurement settings applied to the entangled qubits. When the resulting correlations violate a Bell inequality beyond any local-hidden-variable explanation, the outcomes are proven to be fundamentally unpredictable—not just unknown, but inherently stochastic . This Bell violation effectively "amplifies" the low-grade input randomness into near-perfect private output bits.

The crucial insight is that the Bell test doesn't just confirm entanglement exists; it dynamically probes and certifies the randomness of the quantum measurement process itself .

What this means for digital security

Certifiably perfect randomness removes a foundational vulnerability in cryptographic systems:

• Unconditionally secure keys and signatures become possible when the underlying randomness is mathematically proven unpredictable, even against adversaries with unlimited computational power .
• Zero-trust deployment is the default: the certification holds even if the quantum devices are supplied by an untrusted third party or are suspected of being compromised .
• Public-process integrity improves for lotteries, election audits, and cryptographic sortition, where any hidden bias in the supposedly random process would undermine fair outcomes .

The trade-off is throughput. Achieving perfect certification requires experimental complexity that currently limits the rate at which random bits can be generated, compared to commercial non-certified quantum random number generators.

How ETH Zurich's method compares to JPMorgan's certified randomness expansion

ETH Zurich's announcement in May 2026 arrived just over a year after another important milestone: in March 2025, a team from JPMorganChase, Quantinuum, Argonne National Laboratory, Oak Ridge National Laboratory, and UT Austin demonstrated certified randomness expansion using a 56-qubit trapped-ion quantum computer, also published in Nature . These two achievements represent complementary approaches to the same problem, with different strengths.

ETH Zurich's randomness amplification starts with a large volume of imperfect, public randomness and filters it into a smaller amount of perfect randomness. The technique is device-independent: the mathematical guarantee doesn't depend on trusting the hardware, making it robust even against a malicious device manufacturer . It solves the harder foundational problem—you don't need a trusted perfect seed at all.

JPMorgan's randomness expansion, based on a 2018 protocol proposed by Scott Aaronson, takes a short, trusted random seed and expands it into a much larger volume of certified random output . The experiment used Quantinuum's H2 processor running random circuit sampling and classical verification on exascale supercomputers to certify at least 71,313 bits of entropy . The guarantee is adversarially robust—secure against a malicious quantum computer—but the protocol requires an initial trusted seed that the ETH approach does not .

The two methods address different practical scenarios. JPMorgan's expansion produces substantially more random bits and is closer to integration with existing quantum computing infrastructure . ETH Zurich's amplification solves the seeding problem at a more fundamental level, proving that perfect randomness can be extracted from a world where no trusted randomness exists to begin with .

Neither method is currently a drop-in replacement for standard random number generators in production systems, but together they chart the path from unverifiable statistical randomness—which has always carried an uncomfortable residue of doubt in high-security contexts—toward mathematically certified guarantees. The next challenge will be engineering these proofs-of-concept into hardware and protocols that can operate at scale while preserving their certification guarantees.