LGT put hybrid post-quantum key exchange in front of live online banking customers

A customer signing into LGT's online banking cannot tell whether the session negotiated the post-quantum half of the handshake or used the classical one, and neither can an error dashboard. In testing, Pfister said, some connections failed and others returned to the previous configuration ^[9]. The fallback is what makes the pilot safe to run in production ^[8]. It also means success has to be measured as the share of sessions that actually negotiated ML-KEM, because a downgraded connection still logs the customer in.

Pfister put the failures in a familiar category. Legacy hardware, unsupported firmware and unmanaged libraries are familiar to everyone, he said, and the pilot exposes technical debt that already exists. That is not a reason to wait, in his account; it is why starting early matters for the bank ^[10].

The technical work started before the inventory was finished: a modernized TLS baseline, then the X25519 and ML-KEM pairing in controlled environments ^[7][3]. "If you try to inventory, replace and validate every cryptographic use case at once, it makes the program unmanageable, and you can't measure the outcome," Pfister said ^[4]. On sequencing he was blunter: the post-quantum cryptography migration is not one project and one algorithm change, he said, it is a series of migration waves ^[5].

Quantum safety came up at the firm's annual risk and cyber risk management meeting in 2022 ^[1]. "Quantum safety is a resilience and risk management issue," Pfister said ^[12]. Questions to vendors about product roadmaps, supported algorithms, timelines and compatibility constraints started in 2023 or 2024 ^[6], one to two years after that meeting ^[17]. SiliconANGLE's account of the live pilot is dated September 17, 2026 ^[15], so roughly four years passed between the first risk conversation and one customer-facing system running hybrid keys ^[16].

This is a single interview, given at DigiCert's own event, for which theCUBE is a paid media partner; SiliconANGLE says sponsors have no editorial control over its content ^[14]. The account does not say how much of LGT's cryptographic estate the inventory now covers, or when the later waves land ^[18].

Pfister said the key question for leadership was straightforward: "Which information must remain confidential long enough that it may be exposed to a future cryptographic break?" ^[11] For a team doing this in its own shop, that sorts the queue, along with who ships the change, your team or a supplier. Short-lived data on systems you control is where to run the first wave and absorb the outage. Long-lived data behind firmware you cannot patch puts your date inside a supplier's release schedule, and the only lever there is the contract question about supported algorithms and compatibility constraints ^[6].