System for quantum-safe processing and control of trust-based artificial intelligence in enterprise systems

Abstract

A data processing system for the quantum-safe processing and control of trust-based artificial intelligence in enterprise systems, comprising an input interface for receiving input data, a processing unit for performing inference operations of an artificial intelligence model, a quantum-safe verification unit for cryptographic validation of input data, model states, and output data using quantum-safe signature and key methods, and a hardware-implemented trust execution unit located between the processing unit and an output interface, characterized in that the trust execution unit is configured to calculate a trust value for each output generated by the artificial intelligence model based on several input parameters and to release the output only if a predefined trust threshold is met.where the release of the output is carried out by a physical control instance that operates independently of the processing unit.

Description

Technical field The present invention relates to the field of data processing systems, in particular systems for the secure processing and control of artificial intelligence in enterprise environments. The invention lies at the intersection of artificial intelligence, information processing security technology, and quantum-safe cryptography. Specifically, the invention relates to a device for executing inference processes using trust-based control mechanisms, in which quantum-safe verification of input data, model states, and output data is performed. Furthermore, the invention includes hardware-based control units for evaluating and releasing results generated by artificial intelligence. The technical field also extends to systems for real-time processing, integrity checking of data and models, and secure decision control in distributed and safety-critical enterprise systems. State of the art In the field of modern enterprise systems, artificial intelligence methods are increasingly being used to analyze large volumes of data, generate forecasts, and make automated decisions. Such systems are used particularly in financial applications, industrial controls, medical assistance systems, security platforms, and data-intensive business processes. With the growing use of artificial intelligence, the technical requirements for ensuring the integrity, traceability, and security of the results generated by these systems are also increasing. Data processing systems are known from the state of the art in which artificial intelligence models are used for inference processing in enterprise environments. These known systems process input data, generate outputs, and make them directly available to downstream application layers or user interfaces. However, they often lack an independent technical instance that physically or systemically verifies the trustworthiness of a generated output before its release. Instead, known solutions are predominantly based on purely software-based verification mechanisms, whose depth of control and tamper resistance are limited. Furthermore, cryptographic security mechanisms are known from the state of the art, which protect data during storage or transmission. Such methods include classical signature and key methods, which, however, may only offer limited resilience against future quantum-based attack methods. Moreover, their application in known enterprise systems is mostly limited to communication channels or file systems, but not to the verification of internal model states, inference parameters, or the integrity of results generated by artificial intelligence. Furthermore, systems are known that calculate confidence ratings or risk scores for decisions. These ratings are often based on statistical or rule-based models and are generally processed within the same software environment in which the artificial intelligence runs. This lack of a separate technical execution layer allows for independent control of output release. Consequently, the known solutions have the disadvantage that an erroneous or manipulated AI output may be released despite internal evaluation. Furthermore, logging systems are known that store processing events, output data, or security events. Such logging solutions often serve for subsequent analysis and do not necessarily possess a tamper-proof, quantum-safe structure. In particular, the state of the art often lacks a technical link between logging, trust assessment, model state checking, and output control within an integrated data processing system. Furthermore, hardware-based security modules are known from the prior art that secure key material or execute cryptographic functions. However, these modules are generally not designed to evaluate results generated by artificial intelligence in real time, calculate a confidence score, and, depending on this, release, restrict, or block the output. Likewise, known solutions lack the integration of quantum-safe verification of model states and inference parameters. While the current state of the art offers individual approaches to AI processing, cryptographic security, trust assessment, and logging, it lacks a technically integrated device where a hardware-implemented trust execution unit interacts with a quantum-safe verification unit to release AI outputs only after verifying defined trust parameters. Therefore, there remains a need for a data processing system that overcomes the weaknesses of conventional software-centric security and control approaches and enables secure, quantum-safe, and auditable control of trust-based artificial intelligence in enterprise systems. Object of the invention The present invention is based on the objective of providing a data processing system that enables secure, traceable and controlled processing of artificial intelligence in enterprise systems, thereby overcoming the disadvantages of conventional systems. The specific task is to provide a technical solution in which the results generated by an artificial intelligence model are not immediately output, but are subjected to independent and reliable review before being released. This will involve a hardware-based control unit that assesses the trustworthiness of each individual output and releases it only if predefined criteria are met. Another objective of the invention is to ensure the integrity of input data, model states and inference parameters through the use of quantum-safe cryptography, thus guaranteeing a high level of security even against future attack scenarios. Furthermore, the invention aims to provide a technical solution that enables a combination of trust assessment, quantum-safe verification and tamper-resistant logging in a unified system to ensure reliable decision control in safety-critical and data-intensive business environments. Furthermore, a system should be created that enables adaptive control of expenditures, whereby expenditures are released, restricted or blocked depending on a calculated confidence value in order to avoid incorrect decisions and increase operational reliability. Ultimately, the purpose of the invention is to provide a scalable and technically integrable system that can be incorporated into existing enterprise systems and enables efficient, secure and future-oriented processing of artificial intelligence. Summary of the invention The present invention relates to a data processing system for the quantum-safe processing and control of trust-based artificial intelligence in enterprise systems, in which a secure and controlled execution of inference processes is ensured. The system comprises a processing unit for executing artificial intelligence models, a quantum-safe verification unit for cryptographically checking input data, model states, and output data, and a hardware-implemented trust execution unit, which is arranged between processing and output. The trust execution unit is configured to calculate a trust score for each generated output based on several technical parameters and to release the output only if a defined threshold is met. This prevents unsafe or erroneous results from being passed directly to downstream systems. Furthermore, a logging unit is provided which stores processing flows, trust assessments, and inference paths in a tamper-resistant and quantum-safe structure. The invention thus enables secure, traceable, and adaptive control of AI-based decision-making processes in enterprise environments. Detailed description of the invention The present invention relates to a data processing system for the quantum-safe processing and control of trust-based artificial intelligence in enterprise systems. The system is designed to ensure controlled, secure, and traceable execution of inference processes, whereby each output generated by an artificial intelligence model is checked and evaluated before being passed on. The data processing system comprises an input interface for receiving input data from external sources, a processing unit for performing inference operations on one or more artificial intelligence models, and an output interface for providing the processed results. A hardware-implemented trust execution unit is positioned between the processing unit and the output interface. This unit acts as a physical control instance and operates independently of the processing unit. The processing unit is configured to process input data using an artificial intelligence model and generate output data. Inferential processing is based on defined model parameters and can encompass both deterministic and learning-based decision-making processes. The generated output data is not directly forwarded to the output interface but is first passed to the trusted execution unit. The trust execution unit is configured to calculate a confidence score for each individual output. To this end, the trust execution unit includes a multi-parameter confidence evaluation unit that considers various technical parameters. These parameters include, in particular, the consistency of the input data, temporal characteristics of the processing, deviations in the model state, and historical decision data. The evaluation is performed through a combination of comparison operations, distance calculations, and rule-based evaluations, with the results being aggregated into an overall confidence score. In parallel, a quantum-safe verification unit is provided, which checks the integrity of input data, model states, and output data. This verification unit uses cryptographic methods that are resistant to quantum-based attacks to ensure that the data has not been manipulated during processing. Both static data and dynamic states of the artificial intelligence model are included. Verification takes place before, during, and after inference processing, thus ensuring continuous integrity checking. The trust execution unit is further connected to control logic that governs the release of output data based on the calculated trust value. If a defined threshold is reached or exceeded, the output is released and forwarded to the output interface. If the trust value is below the threshold, the output can either be blocked or provided in a restricted form. This ensures that only results deemed trustworthy are processed further. The system also includes a logging unit that stores all relevant processing information. This includes, in particular, input data, model states, calculated confidence levels, inference paths, and release decisions. Logging is performed in a tamper-proof data structure that also utilizes quantum-safe cryptographic methods. This enables subsequent verification and traceability of all decisions. In a preferred embodiment, the data processing system has a modular design, allowing multiple processing units and trust execution units to be combined. This enables the system performance to be scaled according to the requirements of the respective application. The individual modules are interconnected via secure communication interfaces, with data transmission also protected by quantum-safe methods. In another embodiment, the system is configured to make adaptive adjustments to the confidence thresholds. Historical data and system states are taken into account to dynamically adjust the evaluation criteria. This enables continuous optimization of decision accuracy and increases robustness to varying input data. The present invention thus enables a technical solution in which artificial intelligence is not only used for data processing, but its results are also verified and controlled by an independent, hardware-based control instance. The combination of trust assessment, quantum-safe verification, and tamper-proof logging ensures secure and reliable decision-making in enterprise systems. The invention is particularly suitable for applications with high requirements for security, integrity, and traceability, such as in financial systems, industrial controls, medical applications, or safety-critical infrastructures. The system can be used in both centralized and distributed architectures. It is understood that the embodiments described above are merely exemplary and that various changes and adaptations can be made without leaving the scope of protection of the invention.

Claims

A data processing system for the quantum-safe processing and control of trust-based artificial intelligence in enterprise systems, comprising an input interface for receiving input data, a processing unit for performing inference operations of an artificial intelligence model, a quantum-safe verification unit for cryptographic validation of input data, model states, and output data using quantum-safe signature and key methods, and a hardware-implemented trust execution unit located between the processing unit and an output interface, characterized in that the trust execution unit is configured to calculate a trust value for each output generated by the artificial intelligence model based on several input parameters and to release the output only if a predefined trust threshold is met.where the release of the output is carried out by a physical control instance that operates independently of the processing unit. Data processing system according to claim 1, characterized in that the trust execution unit comprises a multi-parameter trust evaluation unit which calculates the trust value taking into account input consistency, temporal processing parameters, model state deviations and historical decision data. Data processing system according to claim 1, characterized in that a logging unit is provided which stores each output decision, the associated confidence value and an inference path in a quantum-safe, tamper-resistant data structure. Data processing system according to claim 1, characterized in that the trust execution unit has adaptive control logic which selectively releases, restricts or blocks the output depending on the calculated trust value. Data processing system according to claim 1, characterized in that the quantum-safe verification unit is configured to cryptographically check not only the input data but also internal model states and inference parameters, and to release only validated states for further processing.