Systems and methods for AI directed tiered post quantum protection of multimodal data
View Patent ↗Training an artificial intelligence model to categorize data by sensitivity and for applying the model to selectively protect sensitive portions of multimodal datasets. Sensitive training data can be obfuscated with synthetic noise or randomized errors to preserve confidentiality while enabling the model to learn patterns correlated with sensitivity. The trained model is validated on labeled data and can be refined as classification standards evolve. In operation, the classifier assigns sensitivity levels to data elements and directs tiered protection. Elements assigned to a higher relative sensitivity classification level are protected using post-quantum key establishment, for example a key encapsulation mechanism, combined with symmetric authenticated encryption of payloads, and associated metadata is authenticated using a post-quantum digital signature scheme. Less sensitive elements can be protected using conventional symmetric encryption for efficiency. This approach automates sensitivity classification, optimizes cryptographic resource allocation, and improves confidentiality and integrity for simulation and mission data.
1 . A computer-implemented method of training an artificial intelligence, AI, model to categorize data based on sensitivity levels, the method comprising:
a. receiving, via one or more processors, a dataset comprising classified and publicly available data together with for respective data elements of the dataset, extrinsic parameters including metadata attributes stored in a permissioned distributed ledger, usage contexts, access patterns, or interrelations between datasets;
b. obfuscating, via the one or more processors, the classified data by injecting synthetic noise or randomized errors to create a modified training set;
c. training, via the one or more processors, an AI model on the modified training set, the publicly available data, and the extrinsic parameters to identify structural patterns and markers indicative of data sensitivity using the extrinsic parameters without relying on unobfuscated classified content;
d. validating, via the one or more processors, the model's performance using a separate test dataset with known classifications; and
e. generating, via the one or more processors, a sensitivity scoring model configured to accept the extrinsic parameters as inputs and to output a sensitivity score for a data element based at least in part on the extrinsic parameters and to map the score to an ordered set of relative sensitivity classification levels that includes at least a higher relative sensitivity classification level and a lower relative sensitivity classification level.
2 . The method of claim 1 , wherein the obfuscation step preserves the structure of the classified data while altering specific details to maintain confidentiality.
3 . The method of claim 1 , wherein the AI model is a neural network configured with layers optimized for pattern recognition in high-dimensional data.
4 . The method of claim 1 , wherein the validation step includes applying k-fold cross-validation to evaluate the accuracy and reliability of the trained model.
5 . The method of claim 1 , wherein the training incorporates reinforcement learning to adapt the model's categorization rules based on feedback from classification tests.
6 . The method of claim 1 , further comprising the step of refining the model by iteratively adding new training data as classification standards evolve.
7 . The method of claim 1 , wherein the sensitivity scoring model outputs a numerical score s in a bounded range and maps s to the ordered set of relative sensitivity classification levels using configurable thresholds, and wherein a level having a greater threshold is defined as a higher relative sensitivity classification level than a level having a lesser threshold.
8 . A computer-implemented method of applying an artificial intelligence based sensitivity categorization model to military simulations, the method comprising:
a. receiving, via one or more processors, data elements from a military simulation, including operational scenarios and equipment specifications, together with, for respective ones of the data elements, extrinsic parameters including metadata attributes stored in a permissioned distributed ledger, usage contexts, access patterns, or interrelations between datasets;
b. processing, via the one or more processors, the data elements and the extrinsic parameter using a trained sensitivity categorization model to determine sensitivity levels that was trained on a modified training set comprising obfuscated classified data, publicly available data, and the extrinsic parameters and that is configured to determine sensitivity levels using the extrinsic parameters without relying on unobfuscated classified content;
c. classifying, via the one or more processors, each data element by assigning the sensitivity score based at least in part on the extrinsic parameters and mapping the score to the ordered set of relative sensitivity classification levels, including at least a higher relative sensitivity classification level and a lower relative sensitivity classification level;
d. protecting, via the one or more processors, the data elements assigned to the higher relative sensitivity classification level using a post quantum confidentiality mechanism comprising establishing a content encryption key with a key encapsulation mechanism and encrypting payloads using symmetric authenticated encryption, wherein selection of the post quantum confidentiality mechanism is based on the relative sensitivity classification level determined from the extrinsic parameters; and
e. transmitting, via the one or more processors, the protected data to a secure storage system.
9 . The method of claim 8 , wherein the selection of protection is monotone non-decreasing with respect to the relative sensitivity classification level such that a data element at a higher relative sensitivity classification level receives protections that are at least as strong as protections applied to a data element at a lower relative sensitivity classification level.
10 . The method of claim 8 , wherein the military simulation data includes communication logs, mission plans, and asset allocation details.
11 . The method of claim 8 , wherein the post-quantum confidentiality mechanism comprises a key encapsulation mechanism selected from Kyber or a public-key encryption scheme selected from NTRUEncrypt.
12 . The method of claim 8 , further comprising applying a post-quantum digital signature scheme selected from CRYSTALS-Dilithium, Falcon, or SPHINCS+ to authenticate protected data elements and corresponding metadata.
13 . The method of claim 8 , wherein data elements assigned to the lower relative sensitivity classification level are protected using a reduced-rigor protection that omits the post quantum key establishment and applies symmetric authenticated encryption, thereby optimizing computational resources.
14 . The method of claim 8 , further comprising generating a report that logs categorized data elements and their corresponding encryption statuses.
15 . A computer-implemented method of applying a trained sensitivity categorization model to mission and training data, the method comprising:
a. retrieving, via one or more processors, mission or training data together with, for respective data elements of the mission or training data, extrinsic parameters including metadata attributes stored in a permissioned distributed ledger, usage contexts, access patterns, or interrelations between datasets;
b. analyzing, via the one or more processors, the retrieved data using a trained sensitivity categorization model that was trained on a modified training set comprising obfuscated classified data, publicly available data, and the extrinsic parameters and that is configured to assign sensitivity scores using the extrinsic parameters without relying on unobfuscated classified content to assign a sensitivity score to each data element based at least in part on the extrinsic parameters, wherein selection of the post quantum confidentiality mechanism is based on the relative sensitivity classification level determined from the extrinsic parameters and to map each score to an ordered set of relative sensitivity classification levels;
c. categorizing, via the one or more processors, the data according to the ordered set, including at least a higher relative sensitivity classification level and a lower relative sensitivity classification level;
d. protecting, via the one or more processors, the data elements assigned to the higher relative sensitivity classification level using a post quantum confidentiality mechanism comprising establishing a content encryption key with a key encapsulation mechanism and encrypting payloads using symmetric authenticated encryption, wherein selection of the post quantum confidentiality mechanism is based on the relative sensitivity classification level determined from the extrinsic parameters; and
e. storing, via the one or more processors, the protected data in a secured, access-controlled database.
16 . The method of claim 15 , wherein the ordered set of relative sensitivity classification levels comprises at least three levels including a lower level, an intermediate level, and a higher level, and wherein the intermediate level is protected using symmetric authenticated encryption with keys established by a classical key agreement.
17 . The method of claim 15 , wherein the mission data includes operational reports, training assessments, and personnel performance evaluations.
18 . The method of claim 15 , further comprising auditing, via the one or more processors, the stored data periodically to ensure ongoing adherence to classification standards.
19 . The method of claim 15 , wherein real-time processing is implemented to categorize new mission and training data as it is generated or received.
20 . The method of claim 15 , wherein updates to the sensitivity categorization model are propagated automatically across the system to maintain consistency in data classification.