A framework for quantum-secure communications in cyber-physical control systems with experimental demonstration in a nuclear reactor Authors: Konstantinos Gkouliaras, Vasileios Theos, True Miller, Brian Jowers, George Kennedy, Andy Grant, Terry Cronin, Philip G. Evans, Stylianos Chatzidakis Published: 2026 ABSTRACT Quantum key distribution (QKD), a cryptographic method grounded in the laws of quantum mechanics rather than computational hardness, promises provably secure communications. Its integration into critical infrastructure offers a pathway to secure resilient operation of next-generation control systems, from power-intensive data centers to remotely operated microreactors located in energy-deprived remote communities. While QKD has been demonstrated in laboratory and network settings, its use in safety-critical control systems remains unexplored where requirements on latency, key availability, and operational stability pose unique challenges. Here, we introduce a QKD-secured data acquisition and control framework for embedding quantum cybersecurity directly into safety-critical cyber-physical systems and report the first end-to-end experimental demonstration in a nuclear reactor. The framework establishes the conditions necessary to achieve secure, low-latency, real-time operation in safety-critical environments. Using a phase-encoding decoy-state BB84 system deployed on Purdue’s 10 kWth fully digital research reactor (PUR-1) we validate the model and achieve real-time encryption and decryption of 2,000 reactor signals. Experiments demonstrated a stable secret-key rate of 320 kbps with a quantum bit error rate of 3.8% at 54 km, and maximum optical-fiber distances of 82 km with One-Time Pad (OTP) encryption and 140 km with AES-256. We further evaluated compliance with latency and key availability conditions using three cryptographic schemes (OTP, AES-256, ASCON), finding that all variants supported real-time operation at typical data reporting rates. Our results establish a universal framework for quantum-cybersecurity in safety-critical systems, demonstrating that quantum communication technologies can enable secure, low-latency, real-time operation of nuclear reactors and other critical infrastructure. Introduction While a fully digital remotely monitored and operated industrial control system would offer numerous advantages, it could be nevertheless vulnerable to cyber events. The “cyber-attack surface” (i.e., the potentially vulnerable systems and components) would be greater compared to conventional analog designs, as a result of increased automation, complexity, remote operation, and reduction in operational staff1. The industry has been a target of such cyber attacks for the past 30 years, with many attacks aimed at gaining intelligence on Supervisory Control And Data Acquisition (SCADA) networks2. This was ably demonstrated in attacks launched in 2014 against the Korea Hydro and Nuclear Power Co. that specifically targeted the blueprints and electrical flow charts of nuclear reactors3. Independent of the objectives or aims of an attack, attackers almost always first gather information about the target system to identify network topology, software versions, authorization or authentication mechanisms, and critical digital assets. This highlights that the first critical layer of defense against attacks would be to guarantee the confidentiality and authentication of any communication. Although cryptographic implementation in nuclear industrial control systems was first proposed in 20104, practical implementation at an actual power plant has not yet taken place5. Information theoretical security (i.e., unconditional) can be achieved when the well-known symmetric encryption One-Time Pad (OTP) algorithm is implemented. To do so, OTP requires that communication parties have access to continuously refreshed and truly random keys, equal in size to the encrypted data6. These requirements are not easily implemented, as never-used keys need to be distributed between two parties in real time. Instead, proposed cryptographic schemes are based on the computational complexity of public key cryptography, i.e., on the difficulty of reversing certain mathematical functions. Unfortunately, public key cryptography exhibits high latency and requires significant memory and processing resources, a limitation especially for safety-critical control systems where high performance computing resources are not readily available. Although the required resources are dependent on the specific implementation, it is found that public key encryption is approximately 2–3 orders of magnitude slower compared to symmetric encryption, comes with considerably higher computation costs, while the several thousands of cryptographic operations required per second could be difficult to manage even for desktop computers6,7. As an example, a microcontroller implementation of Elliptic Curve Cryptography (ECC) for key exchange and authentication required approximately 28 kB more flash memory and 8 kB more heap RAM compared to using pre-shared keys, while consuming over 33 times more energy8. Even more, public key cryptography has been shown to be vulnerable to the advent of quantum computers and Shor’s algorithm, and to “store now, decrypt later” strategies9–11. For instance, predictions on the evolution and scalability of quantum computers reveal that they could compromise public key encryption (RSA) within hours12. Large scale quantum computers in the next decade is a realistic expectation with several initiatives launched, including the National Quantum Initiative Act13, Google’s AI Quantum Laboratory with plans to commercialize quantum computers14–16, IBM Q17,18, etc. A promising approach to address this challenge is Quantum Key Distribution (QKD). QKD relies on the inherent properties of quantum mechanics to deliver unconditional security, without making assumptions on adversarial resources or strategy. QKD is the most mature quantum cryptography application, enabling secure generation and distribution of a truly random key at two distant locations19. Since its first introduction in 1984 with the BB84 protocol20, there have been multiple protocols21–25, security proofs26–34, hardware advancements35–37, experimental network demonstrations38–48, and commercial realizations49. Current research efforts are also focused on identifying potential security challenges in real-world deployment due to imperfect hardware (e.g., single-photon detectors)19. Usage of practical equipment could lead to deviations from the protocol’s information-theoretic security, as claimed for the case of the Micius quantum satellite50. In addition to QKD, quantum secure direct communication (QSDC) has been proposed and investigated as an alternative quantum communication protocol, leveraging quantum physics at the stage of message transmission as opposed to the stage of key distribution51. Quantum communication networks can be further coupled with post-quantum digital signatures for access authentication, as shown in52. QKD has received significant attention for critical infrastructure and energy sector53. Proposed application areas include the smart grid54–58, microgrids59, electrical utilities60, and hydropower plants61. A detailed review of QKD energy applications is found in53. However, examining QKD potential for the nuclear power sector is a particularly underexplored topic, with related work limited in a preliminary concept study62, an investigation of wireless QKD application63, and a study on atmospheric optical communication64. Previous work included the development of a QKD simulator65, leveraged to evaluate QKD performance in different nuclear environments66. In this paper, we present a quantum cybersecurity framework, derived from first principles and generalizable to any industrial control system, for embedding quantum-secure communications directly into data acquisition and control. The framework establishes the conditions necessary to achieve secure, low-latency, real-time operation in safety-critical environments. To validate the model, we perform the first complete end-to-end demonstration of a phase-encoding, decoy-state BB84 QKD protocol deployed under prototypic operating conditions on Purdue’s fully digital research reactor, PUR-1. In our implementation, the Toshiba Long-Distance QKD system (QKD-LD) was coupled with PUR-1 to enable encrypted transmission of live reactor signals, providing an experimental platform to test the model’s requirements in practice. An extensive series of measurements were conducted to evaluate key generation rates, quantum bit error rates, and system latency across different fiber transmission distances, operational use cases, and encryption algorithms. QKD-secured industrial control system communication model Industrial control systems have different operational requirements than conventional communication or network systems. Any encryption must satisfy two characteristics that are particularly critical. First, they must achieve practically 100% system availability: interruptions to operation are unacceptable except during scheduled and absolutely necessary tasks (e.g., refueling). Unlike networked communication systems, where temporary outages or retransmissions can often be tolerated, safety-critical control loops cannot risk downtime. Second, latency and continuity requirements are absolute. No data gaps or discontinuities can be permitted, since the data historian must precisely archive past states to guarantee accurate reconstruction and traceability of system behavior. To evaluate the behavior of QKD in an industrial control system, we develop a model that captures these constraints. We define several key input parameters governing data generation, encryption, and transmission, and characterize the different stages of the data exchange loop including the conditions necessary for sustaining uninterrupted, low-latency secure communication. Communication workflow A generic QKD system consists of two channels (Fig. 1): a quantum channel (optical fiber or free space) and a classical channel (data link). The quantum channel is used to exchange single photons which carry information encoded in one of their degrees of freedom, such as polarization or phase. Once the photon pulses are detected, a series of communication rounds occur via the classical channel to correct errors and remove any knowledge potentially leaked to an adversary. The two parties gain access to quantum-random bit strings, which can be used with symmetric OTP encryption to provide unconditional security. While QKD has been demonstrated extensively in laboratory and communication network settings, integration into safety-critical control systems introduces fundamentally different constraints. Unlike in conventional network applications, where QKD is used to secure point-to-point or multi-user links, integration into control systems requires mapping the cryptographic workflow directly onto real-time data acquisition and actuation loops. Fig. 1QKD key distillation procedure as presented in67. QKD key distillation procedure as presented in67. Figure 2 illustrates this process in the context of a digital control system. Terminal A fetches data from the digital controller and encodes them in a digital representation scheme (e.g., IEEE-754). Then, it communicates with Key Server Alice requesting an encryption key, the size of which is dictated by the encryption algorithm and/or data size. Key Server Alice replies by sending the key and the associated key ID. The data are encrypted, and the ciphertext is transmitted to Terminal B over the authenticated channel along with the key ID. Upon receipt, Terminal B requests the decryption key from Key Server Bob, providing the ID. Communication between servers and terminals is authenticated to ensure that only the legitimate parties can gain access to the symmetric keys. The ciphertext is decrypted and data are extracted. Finally, an automated model analyzes the received data and provides some form of feedback to the operator regarding action to be taken. For example, the model might provide guidance for rod movement to match the target reactor power or neutron flux. All communication between servers and terminals is authenticated to ensure that only legitimate parties gain access to symmetric keys. This closed-loop architecture provides the foundation for evaluating the feasibility of QKD-secured control system operation. Fig. 2Schematic of data communication loop. Schematic of data communication loop. Input parameters for communication workflow To model this workflow, we introduce input parameters grouped into data, security, and quantum channel parameters. The input parameters are summarized in Table 1. Data parameters determine how much information must be encrypted and transmitted; security parameters define how key material is consumed and how latency accumulates across the workflow; quantum channel parameters dictate how quickly key material can be supplied by the QKD system. Data parameters Data parameters are associated with signal resolution and quantity. These include the number of signals (N), precision (p), data sampling rate (fs) which describes the frequency with which the controller records field sensor data, and data reporting rate (frep) which is a quantity referring to the rate at which data are reported to the remote terminal unit. The data sampling rate and data reporting rate do not need to be identical by default. For example, data could be sampled every 1 millisecond but reported once every 100 milliseconds. In this scenario, intermediate values are discarded. However, this might not always be the case. It could be possible that even though data are updated 10 times per second (fs = 10 Hz), all values are reported once every second as a batch (frep = 1 Hz). Precision is the number of bits required to encode each data point in digital format and is dependent on the data representation scheme, making assumptions on the floating-point accuracy. Using the IEEE-754 standard for floating point arithmetic68, the main options are either single precision (32 bits per value) or double precision (64 bits per value). For each reporting period, the amount of data generated in bits is:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{n}_{{\Delta}{\tau}_{ef}}=N\cdot\:p\cdot\:{f}_{s}\cdot\:{\Delta}{\tau}_{ef},$$\end{document} where Δτef is the effective data period:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\Delta\:}{\tau\:}_{ef}=max\left(\frac{1}{{f}_{s}},\frac{1}{{f}_{rep}}\right).$$\end{document} Security parameters Security input parameters are needed to model compliance with confidentiality and availability requirements. These include the key reusability factor (fenc) and elapsed times associated with each communication stage. The key reusability factor describes the number of disposable key bits required to encrypt one bit of data and depends on the cryptographic algorithm implemented, i.e.,3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{f}_{enc}=\frac{key\:size}{encrypted\:data\:size}\in\:\left(\mathrm{0,1}\right]$$\end{document} Lower values of the ratio represent higher key economy, potentially at the cost of theoretic security. Information-theoretic security requires one disposable key bit per encrypted data bit, thus fenc = 1, which is the case with OTP. All remaining symmetric cryptographic algorithms practically exhibit fenc < 1, as key material is somehow reused. In those cases, coupling with QKD offers only computational security. However, security bounds can be improved by increasing the refresh rate of a truly random key, a task which is challenging when using conventional key distribution methods (e.g., public cryptography). A QKD implementation has the potential to actually upgrade communication confidentiality, not only in the ideal scenario of coupling with OTP, but even when combined with applied encryption algorithms. The number of key bits needed per effective period is:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:n\left(N,p,{f}_{s},{{f}_{rep},\:f}_{enc}\right)={n}_{{\Delta\:}{\tau}_{ef}}\cdot\:{f}_{enc},$$\end{document} and latency:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:L={t}_{fetch,A}+{t}_{key,\:A}+{t}_{enc,A}+{t}_{transm}+{t}_{key,\:B}+{t}_{dec,B}+{t}_{action,B},$$\end{document} where each time variable represents the elapsed time associated with each communication stage. Indexes A and B represent the terminal where each process occurs. Quantum channel parameters Quantum channel input parameters are related to the properties of the QKD equipment and the quantum transmission link. These include channel length (l), channel attenuation coefficient (α), quantum bit error rate (QBER), sifting ratio (ηsift), error correction efficiency (fec), photon source repetition rate (fsource), and detector efficiency (ηd). Channel length is the distance between the reactor site and the remote location. Along with the quantum bit error rate (channel noise), it determines the secret key generation rate, as larger distances introduce increased photon attenuation in the optical fiber. The secret key rate is67:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:SKR=\:{R}_{raw}\cdot\:{\eta}_{sift}\cdot\:g\left(E\right),$$\end{document} where Rraw is the raw key rate, ηsift is the fraction of raw key bits remaining after discarding measurements conducted with a different basis than the one used for encoding, and g(E) is a function that depends on the quantum bit error rate and describes the inconsistencies between the sifted bit streams held by the two parties. While the sifting ratio for the original QKD protocol was 50%, asymmetric implementations allow to improve efficiency through biased basis selection. The analytical forms of Rraw and g(E) depend on the specific QKD protocol and postprocessing algorithms implemented. In a qubit-based system, the two terms are given as67:7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}_{raw}={f}_{source}\cdot\:{\eta}_{d}\cdot\:{t}_{chan},$$\end{document} where tchan is the channel transmissivity, given as a function of channel attenuation coefficient and channel length:8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{t}_{chan}={10}^{-al/10}.$$\end{document} The function g is:9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:g\left(E,\:\overline{{e}_{1}},\:\underline{{p}_{1}}\right)=\underline{{p}_{1}}\left[1-h\left(\overline{{e}_{1}}\right)\right]-{f}_{ec}h\left(E\right),$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\underline{{p}_{1}}$$\end{document} is the lower bound for the probability that a sifted bit was registered from the detection of a single photon state, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\overline{{e}_{1}}$$\end{document} is the upper bound for error rate in single photon states, h is the binary entropy function, and E is the QBER. The two bounds are estimated as dictated by the specific QKD protocol used67. A milestone in practical QKD realization was the development of the decoy state technique69–71. In decoy state protocols, signal photon states are randomly blended with the so-called decoy states to achieve higher key rates and security against advanced eavesdropping strategies, such as the Photon-Number Splitting (PNS) attack72. An important contribution of the decoy state technique is the ability to approach the security of ideal QKD protocols while using practical, imperfect hardware. An alternative approach for PNS-robustness avoiding decoy states would be the use of single-photon emitters; despite traditional hardware limitations, a recent experimental demonstration showed the potential to achieve secure key generation leveraging a GaN defect-based source at room temperature, compatible with the telecom band73. System requirements Integration of QKD into a safety-critical control system imposes two primary requirements: key availability and latency. Both conditions must be satisfied simultaneously for a given configuration to be realizable. Key availability The QKD system must generate secret key material at least as quickly as it is consumed for encryption. The target key availability condition dictates that the secret key formed during the data generation interval should be at least equal in size to the key n required to encrypt the data. This requirement can be expressed as:10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:n\left(N,p,{f}_{s},{f}_{rep},{f}_{enc}\right)\le\:SKR\left(l,E,\:a,\:{\eta}_{d},{\eta\:}_{sift},\:{f}_{source},\:{f}_{ec}\right)\cdot\:{\Delta\:}{\tau}_{ef},$$\end{document} where l and E are the distance and error rate of the quantum channel, respectively. SKR is the corresponding secret key rate reported in the effective period Δτef, and is dependent on the quantum channel parameters. The required key size is a function of the data and security parameters. Table 1Summary of communication workflow input parameters.SymbolNameUnitsDescriptionData parameters N Number of signals–Number of signals to be transmitted f s Sampling rates−1Rate of controller signal update f rep Reporting rates−1Rate at which signals are sent to remote location p PrecisionbitsNumber of bits allocated to each valueSecurity parameters f enc Key reusability factor–Disposable key bits required per bit of encrypted datatkey, tfetch, tenc, ttransm, tdec, tactionElapsed timesElapsed time associated with each communication stage or terminalQuantum channel parameterslChannel lengthkmDistance between two secure locationsEQuantum bit error rate%Fraction of bits corrupted by noise/errors in the quantum channel α Attenuation coefficientdB/kmSignal loss per unit length in the quantum channel η d Detection efficiency–Probability that a photon arriving at the detector is successfully detected η sift Sifting ratio–Fraction of raw key bits retained after basis reconciliation f source Source repetition rates−1Rate at which quantum signals are emitted by the source f ec Error correction efficiency–Efficiency factor of the error correction protocol Table 1Summary of communication workflow input parameters.SymbolNameUnitsDescriptionData parameters N Number of signals–Number of signals to be transmitted f s Sampling rates−1Rate of controller signal update f rep Reporting rates−1Rate at which signals are sent to remote location p PrecisionbitsNumber of bits allocated to each valueSecurity parameters f enc Key reusability factor–Disposable key bits required per bit of encrypted datatkey, tfetch, tenc, ttransm, tdec, tactionElapsed timesElapsed time associated with each communication stage or terminalQuantum channel parameterslChannel lengthkmDistance between two secure locationsEQuantum bit error rate%Fraction of bits corrupted by noise/errors in the quantum channel α Attenuation coefficientdB/kmSignal loss per unit length in the quantum channel η d Detection efficiency–Probability that a photon arriving at the detector is successfully detected η sift Sifting ratio–Fraction of raw key bits retained after basis reconciliation f source Source repetition rates−1Rate at which quantum signals are emitted by the source f ec Error correction efficiency–Efficiency factor of the error correction protocol Summary of communication workflow input parameters. Latency The second requirement concerns end-to-end communication latency. Since data are transmitted to a remote terminal in fixed time intervals, the complete encryption, transmission, and decryption loop must be completed within this interval. In scenarios where data are batched, the threshold is instead set by the reporting period:11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{t}_{fetch,A}+{t}_{key,\:A}+{t}_{enc,A}+{t}_{transm}+{t}_{key,\:B}+{t}_{dec,B}+{t}_{action,B}\le\:max\left(\frac{1}{{f}_{s}},\frac{1}{{f}_{rep}}\right).$$\end{document} Each time variable represents the elapsed time associated with each communication stage, with respect to Fig. 2. Neglecting the time required for data analysis, this can be compactly expressed as:12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{t}_{total}={t}_{qkd}+{t}_{crypto}+{t}_{comm}\le\:max\left(\frac{1}{{f}_{s}},\frac{1}{{f}_{rep}}\right),$$\end{document} where tqkd = tkey, A + tkey, B is the total time involving key request and delivery via QKD server communication, tcrypto = tenc + tdec is the encryption/decryption interval, and tcom = tfetch, A + ttransm is the overall time required for data exchange (fetching from PLC and transmitting between Terminals A and B). Together, key availability and latency form the two principal performance metrics for QKD-secured control systems. For a given use case, the data and channel parameters define the target values for both metrics. A configuration is feasible only if both conditions are simultaneously satisfied. Experimental setup A prototypic QKD setup was installed in PUR-1 reactor. PUR-1 is a pool-type research reactor located at Purdue University (Fig. 3). To date, it is the only fully digital reactor licensed by the US NRC. The Programmable Logic Controller (PLC) allows remote monitoring and collection of more than 2,000 parameters, including digital values (e.g., manual SCRAM control) and digitized analog quantities (e.g., neutron flux). Of these, 67 signals have been found to be most relevant for representing important system behavior. This selection has been conducted based on domain knowledge, excluding parameters not directly related to the power generation process (e.g., room temperature)74. A timestamp is also included to form a set of 68 signals in total. Fig. 3PUR-1 reactor with remote OT/IT anomaly monitoring workstation. PUR-1 reactor with remote OT/IT anomaly monitoring workstation. The experimental setup, shown in Fig. 4, uses the commercial Toshiba Long Distance QKD system (QKD-LD). QKD-LD consists of four devices: QKD-Alice, QKD-Bob, Key Server Alice and Key Server Bob. QKD-Alice and QKD-Bob connect using a quantum and a classical channel, both of which are single-mode optical fibers. For the quantum communication stage, QKD-LD implements the T12 QKD protocol67,75. T12 is a modified version of phase-encoding decoy state BB84, using asymmetric basis selection. One decoy and a vacuum state are blended with signal states with probabilities of 1.661% and 1.466%, respectively. The prototypic QKD testbed as installed in PUR-1 reactor is shown in Fig. 5. The system is structured around two workstations for the sender (Alice) and receiver (Bob), referred to as WA and WB. WA and WB represent the two remote locations, with WA physically connected to PUR-1. WA and WB run Windows 11 OS and communicate over a regular TCP/IP, non-dedicated LAN connection, shared with other network devices to perform encryption and decryption operations on real-time reactor data. WA and WB are connected to the QKD servers of Alice and Bob, respectively, over dedicated Ethernet CAT-5e data links. The servers, running Linux OS, are responsible for implementing the Key Management System (KMS) by storing the QKD-generated keys and providing them upon request to the workstations, according to the ETSI GS QKD 014 standard76. Whenever one of the two workstations requests a key, the corresponding KMS server replies with a unique key ID associated with the key. The key ID, a 128-bit Universally Unique Identifier (UUID), is simultaneously forwarded to the second server. The remote workstation receives the key ID over an authenticated channel and uses it to request the corresponding key from the KMS server. Through this process, the two communicating parties gain access to an identical key to subsequently use with a symmetric encryption scheme. Fig. 4Schematic of PUR-1 QKD experiment setup. Schematic of PUR-1 QKD experiment setup. The connections between workstations and servers are authenticated through the mutual Transport Layer Security (mTLS) protocol, using signed certificates. The server verifies the validity of the certificate in possession of the key requester, thus confirming their identity before key delivery is allowed. The mutual authentication step—taking place in every key delivery iteration—allows one to verify that only the legitimate parties can have access to the keys, a fundamental assumption of the provided security76. Since a single key cannot be requested twice and the communication between the workstation and the server is authenticated, the key ID does not need to be encrypted. The two QKD devices (QKD-Alice, QKD-Bob) are connected to Key Server Alice and Key Server Bob through duplex multi-mode fiber. Following the QKD protocol specification, they are connected to each other through a quantum and a classical channel. Both channels are single mode fibers (Corning SMF-28) terminated with LC/UPC connectors. To experimentally replicate different transmission distances and environmental conditions, additional equipment intercepts both SMF channels. Three devices are connected sequentially, a General Photonics Polarization Controller, a DiCon Variable Optical Attenuator (VOA), and a DiCon GP800 delay line. Figure 5 shows the stack of equipment intercepting the QKD channels. The polarization controller enables manual control of photon polarization by defining a voltage applied to a series of fiber squeezers. The VOA introduces channel attenuation up to 30 dB per channel to replicate longer transmission distances. In addition, two fixed 10 dB attenuators are introduced for the classical and quantum channel. The delay line uses MicroElectroMechanical (MEMS) optical switches to vary the fiber length connected, ranging from 0 to 32 km, introducing not only photon attenuation but potential time/frequency dispersion effects. Typical loss in SMF operating at a wavelength of 1550 nm is approximately 0.2 dB/km77, thus photon attenuation of a channel up to 200 km can be replicated. Combining the VOA and delay line allows for a more realistic fiber implementation while also further increasing the channel length capability. The TOSHIBA system reports the SKR and QBER, along with the corresponding timestamp. QKD data are reported after each key distillation cycle is completed. Fig. 5QKD testbed in PUR-1 control room (Left: QKD stack; Middle: Encryption/decryption workstations; Right: Equipment to adjust quantum channel parameters). QKD testbed in PUR-1 control room (Left: QKD stack; Middle: Encryption/decryption workstations; Right: Equipment to adjust quantum channel parameters). Results The analytical framework developed in sections “System requirements” and “Experimental setup” established two fundamental feasibility conditions for QKD-secured reactor communication: key availability (Eq. 10) and latency (Eq. 12). These define the operational envelope within which encrypted remote monitoring and control can occur without violating safety-critical requirements. In this section, we validate the model using experimental data obtained in PUR-1 with the Toshiba QKD LD system. The QKD setup was operated at various channel lengths by adjusting VOA and delay line parameters, and sustained measurements were collected for at least 10 h at each configuration. For each case, the SKR and QBER were recorded and compared against the thresholds predicted by the model. The measured SKRs were then applied in the key availability condition to determine feasible distances for representative use cases, while cryptographic timing experiments for multiple cryptographic schemes (OTP, AES-256, ASCON) were used to evaluate compliance with the latency condition and analyze their ability to satisfy control-system constraints. SKR and QBER Measured SKR and QBER values are shown in Fig. 6. We found an average rate of approximately 315 kbps at 50 km, decreasing exponentially as the channel length increases. At 145 km, no secret key can be distilled, leading to zero secret key rate. Performance aligns with the 30 dB loss design of the Toshiba QKD LD system. QBER ranges from approximately 4% at 50 km to 7.5% at 145 km. The curve shows that, despite channel length and QBER not being directly related, increasing distance contributes to higher error rate in practice. The total absence of key generation at 145 km is thus attributed not only to increased photon attenuation, but also to high error rate on the pulses detected, which further prevents secure key distillation. The variance is approximately 10.3 kbps for SKR and 0.07% for the error rate, at a 54 km distance. The measured SKRs and QBERs serve as direct inputs to the key availability condition and define the maximum feasible reactor workloads at different distances discussed in section “Key availability”. Fig. 6Average SKR and QBER as a function of distance. QKD data collected over 10 h of operation at each length (left). SKR and QBER versus time for 10 h of operation at l = 54 km (right). Average SKR and QBER as a function of distance. QKD data collected over 10 h of operation at each length (left). SKR and QBER versus time for 10 h of operation at l = 54 km (right). Key availability The key availability condition (Eq. 10) requires that the amount of secret key generated during a reporting period is at least as large as the key size demanded by the encryption scheme. In practice, this sets a distance-dependent limit on the number of reactor signals that can be transmitted securely at a given sampling frequency and cryptographic configuration. Figure 7 overlays the measured SKRs against the calculated key consumption rates for representative use cases. Assuming conservative OTP encryption (fenc = 1) with fs = frep, the model predicts that secure operation is feasible up to ~ 85 km for 2,000 signals. When only 68 signals are transmitted, the model tight bound increases to ~ 135 km at fs = 1 Hz and ~ 105 km at fs = 10 Hz. Experimental SKRs align closely with these boundaries, confirming the predictive accuracy of the model. We further benchmarked AES-256 in CBC mode, which requires a fixed 256-bit key and 128-bit IV per block, regardless of the number of signals. The model indicates that this key demand translates to a constant bandwidth requirement dependent only on fs, not on N. Measurements confirm that AES extends feasible distances to ~ 140 km, demonstrating how the model generalizes across different cryptographic families. Latency The latency condition (Eq. 12) constrains the total processing and transmission time per reporting interval. It requires that the combined durations of QKD key exchange, cryptographic operations, and data transfer remain below max(1/fs, 1/frep). This condition ensures that reactor data are securely transmitted without introducing discontinuities, while preserving plant availability. Latency experiments were conducted for OTP, AES-256, and ASCON variants. All 2,000 signals are fetched, encrypted, and transmitted according to the communication procedure described in section “QKD secured industrial control system communication model”. The experiment is conducted for 15,000 data generation cycles, at a rate of 10 samples/sec. Figure 8 shows the total latency as well as those corresponding to each module, for OTP and AES. For OTP with 2,000 signals at fs = 10 Hz, the model predicts that the reporting interval (100 ms) is the critical threshold. Similarly, the critical threshold increases to 1 s for fs = 1 Hz. The average time per data cycle is 395 ± 15 ms, out of which 248 ms and 145 ms are reported in the QKD and Crypto modules, respectively. Fig. 7Maximum distance satisfying target key availability condition for various communication scenarios. Maximum distance satisfying target key availability condition for various communication scenarios. The experiment was repeated again for 68 signals. Although there is significant difference in terms of key availability, the number of signals does not considerably affect latency in OTP encryption with the Crypto module requiring approximately 50 ms more. AES-256 is the most secure variant of the Advanced Encryption Standard, featuring a 256-bit key and 128-bit IV. Since both the key and the IV are updated per data cycle, a total of 384 bits are requested at every iteration. It is noted that the IV does not need to be secret in general to achieve the algorithm’s computational security. Nevertheless, our implementation allocates QKD-generated material for the IV to ensure that the sequence is truly uniform6 and to provide a potentially tighter security limit. In addition, assigning additional secret bits allows one to generalize the latency results for cryptographic variants involving longer keys. Using Eq. 3, the key reusability factor is found equal to 17% and 0.6% for N = 68 and N = 2, 000, respectively. Although AES has higher complexity compared to the bitwise operations in OTP, the results do not indicate additional delay in our implementation. A similar approach was followed to evaluate the performance of ASCON variants. ASCON is the official selection of NIST for the novel field of Light-Weight Cryptography (LWC). LWC algorithms are designed around simpler operations, to enable cryptographic capabilities even in devices with limited computational resources78,79. In this implementation, three prominent variants are examined (ASCON-128, ASCON-128a, ASCON-80pq). Notably, ASCON-80pq is a post-quantum cryptography cipher, claiming to be more robust against a quantum computer-initiated attack80. The ASCON family offers authenticated encryption with associated data (AEAD). The key size is 128 bits for the 128 and 128a variants, while 80pq handles 160-bit key. All three variants further require a 128-bit nonce. As in the previous cases, the key and nonce are refreshed per data generation cycle. The key reusability factor is 0.4% for ASCON-128/128a and 0.45% for ASCON-80pq, when N = 2000. Results using ASCON variants are shown in Fig. 9. All three variants remain within the boundaries of 1 and 10 sampling iterations per second. While there is still delay for a use case of 10 Hz, the cryptographic module time is slightly reduced compared to OTP and AES by approximately 20–25 ms per cryptographic cycle. The QKD module intervals continue to average around 245 ms, confirming that the elapsed time for establishing communication between QKD servers and secure applications includes an unavoidable overhead. Fig. 8Latency for OTP and AES-256 encryption. Latency for OTP and AES-256 encryption. Fig. 9Latency for ASCON variants (N = 2000 signals, p = 32 bits). Latency for ASCON variants (N = 2000 signals, p = 32 bits). The performance metrics for all algorithms and configurations are summarized in Table 2 and displayed in Fig. 10. Overall, the total latency is consistently within the acceptable bounds, for the use case where data is sampled every second. There is an average margin of more than 0.5 s to immediately compensate for any delay outliers (e.g., the two peaks of Fig. 16a). Together, these results confirm that the latency condition provides a robust predictor of feasible configurations. The experiments highlight that QKD server communication introduces a fixed overhead (~ 245 ms), while cryptographic choices determine whether this overhead remains compatible with sampling-rate requirements. Table 2Latency metrics for different encryption algorithm variants and number of signalsAlgorithmSignalsTotal time (ms)QKD time (ms)Crypto time (ms)MeanSTDMeanSTDMeanSTDOTP68444.9520.91248.7617.28195.7911.47OTP2,000395.2314.61248.1811.32144.728.33AES-25668440.4716.66244.7911.70195.3210.83AES-2562,000390.4514.71244.3211.48144.277.86ASCON-1282,000419.9114.49243.0511.16175.288.00ASCON-128a2,000417.2320.43245.7016.42169.9310.42ASCON-80pq2,000421.5327.83244.0226.11175.898.30 Table 2Latency metrics for different encryption algorithm variants and number of signalsAlgorithmSignalsTotal time (ms)QKD time (ms)Crypto time (ms)MeanSTDMeanSTDMeanSTDOTP68444.9520.91248.7617.28195.7911.47OTP2,000395.2314.61248.1811.32144.728.33AES-25668440.4716.66244.7911.70195.3210.83AES-2562,000390.4514.71244.3211.48144.277.86ASCON-1282,000419.9114.49243.0511.16175.288.00ASCON-128a2,000417.2320.43245.7016.42169.9310.42ASCON-80pq2,000421.5327.83244.0226.11175.898.30 Latency metrics for different encryption algorithm variants and number of signals Fig. 10Average latency for OTP, AES, and ASCON cryptographic variants. Average latency for OTP, AES, and ASCON cryptographic variants. Conclusions In this work, we have presented the first end-to-end experimental demonstration of a QKD-secured data acquisition and control framework in a nuclear reactor. By embedding a phase-encoded decoy-state BB84 system in PUR-1, we demonstrated secure real-time encryption and decryption of 2,000 reactor signals across optical fiber links up to 140 km. The integration of QKD with OTP, AES, and ASCON encryption validated the feasibility of maintaining low-latency, high-availability communication under safety-critical conditions. We proposed a control system specific performance model for QKD-secured communication. Unlike conventional network applications, control systems impose stringent requirements on availability, continuity, and latency. The model formalized these constraints through key availability and latency conditions, allowing us to derive feasibility limits for different cryptographic schemes and sampling configurations. Experimental results closely matched model predictions, confirming that the proposed framework captures the fundamental operational trade-offs. This combination of model-driven analysis and experimental validation provides a universal methodology for assessing the integration of quantum cybersecurity into cyber-physical systems. Our findings highlight the potential of quantum-secure communications for safety-critical energy systems. By demonstrating that QKD can sustain secure, low-latency, real-time reactor operation, this work provides both a foundational framework and a proof-of-principle demonstration that quantum communication technologies can play a central role in enabling resilient, secure autonomous operation of next-generation energy systems.