Detecting anomalous resource distribution patterns in distributed artificial intelligence-based agent networks

Patent metadata

Field	Value
Publication number	US-12592897-B2
Application number	US-202519333687-A
Country	US
Kind code	B2
Filing date	Sep 19, 2025
Priority date	Jan 19, 2023
Publication date	Mar 31, 2026
Grant date	Mar 31, 2026

Abstract

Systems and methods disclosed herein automatically detect, analyze, and mitigate anomalous resource distribution among artificial intelligence (AI)-based agents within a distributed computational network. The system receives a resource allocation request specifying computational resources, agent parameters, and performance objectives for a network of agents. A first AI model set monitors agent activity by tracking resource consumption and behavioral deviations from baseline profiles. The system compares resource usage of agents with historical norms and/or predetermined thresholds to generate an anomaly score for each agent. A second AI model set aggregates scores to construct a multi-dimensional data structure that indicates the comparison and anomaly score. The system ranks agents by anomaly severity to isolate agents with high scores (e.g., misaligned agents), and reallocates resources and updates access privileges for the misaligned agents.

First claim

We claim:

1 . A system for detecting anomalous resource distribution in distributed artificial intelligence (AI)-based agent networks, the system comprising:
at least one hardware processor; and at least one non-transitory memory storing instructions, which, when executed by the at least one hardware processor, cause the system to:
access, from a distributed ledger network, operational data for each of multiple AI-based agents that defines (a) resource allocation traffic data collected during execution of one or more tasks autonomously executed by the AI-based agent and (b) an expected resource allocation data structure that defines one or more allocation thresholds for resources allocated by the AI-based agent,
wherein the resource allocation traffic data and the expected resource allocation data structure are stored as records within the distributed ledger network;

route the operational data for each of the multiple AI-based agents to a subset of a validator AI model set by:
determining a dimension set of the operational data against which to evaluate the operational data, and
for each particular dimension in the dimension set, identifying the validator AI model subset configured to test the particular dimension, wherein each validator AI model within the validator AI model set is trained to test a different dimension;

identify, using the validator AI model subset, an anomalous agent set within the multiple AI-based agents by:
comparing, for each AI-based agent, the resource allocation traffic data against the one or more allocation thresholds defined in the expected resource allocation data structure; and
determining, for each anomalous agent, that the resource allocation traffic data fails to satisfy the one or more allocation thresholds;

in response to identifying the anomalous agent set, generate a sequence of computer-executable instructions configured to automatically block transmittal of additional traffic data to each anomalous agent; and
store, in the distributed ledger network, a record set that includes a representation of the validator AI model subset, the sequence of computer-executable instructions, the anomalous agent set, and a timestamp of the identification of the anomalous agent set.

2 . The system of claim 1 , wherein the system is further caused to:
access a network topology data structure that indicates communication between the multiple AI-based agents,
wherein the validator AI model subset is configured to access the expected resource allocation data structure by filtering the records within the distributed ledger network using the network topology data structure.

3 . The system of claim 1 ,
wherein the resource allocation traffic data comprises one or more transaction records stored across multiple blockchain nodes, and wherein each transaction record includes one or more of: an agent identifier, a timestamp, or a token usage amount.

4 . The system of claim 1 ,
wherein the one or more allocation thresholds correspond to one or more anomaly indicators, and wherein the one or more anomaly indicators include one or more of: a network communication pattern, a reputation score, a task completion rate, a response time, a price threshold, or a resource usage amount.

5 . The system of claim 4 , wherein the system is further caused to:
in response to identifying the anomalous agent set, adjust a reputation score associated with each anomalous agent,
wherein the reputation score indicates a degree of satisfaction of the resource allocation traffic data with the one or more allocation thresholds.

6 . The system of claim 1 ,
wherein the validator AI model subset includes at least one domain-specific model, wherein the operational data for each of the multiple AI-based agents is routed to the validator AI model subset trained on data sharing a common domain with one or more AI-based agents, and wherein the domain indicates one or more of: 1) an area of knowledge, 2) a data type, or 3) a type of task.

7 . A non-transitory, computer-readable storage medium comprising instructions thereon, wherein the instructions, when executed by at least one data processor of a system, cause the system to:
access, from a federated database, operational data for each of multiple AI-based agents that defines (a) resource allocation data collected during execution of one or more tasks executed by the AI-based agent and (b) an expected resource allocation data structure that defines one or more allocation thresholds for resources allocated by the AI-based agent,
wherein the resource allocation data and the expected resource allocation data structure are stored as records within the federated database;
route the operational data for each of the multiple AI-based agents to a subset of a validator AI model set using a dimension set of the operational data against which to evaluate the operational data,
wherein the validator AI model set is trained to test different dimensions;
identify, using the validator AI model subset, an anomalous agent set within the multiple AI-based agents by comparing, for each AI-based agent, the resource allocation data against the one or more allocation thresholds defined in the expected resource allocation data structure; in response to identifying the anomalous agent set, generate a series of computer-executable instructions configured to block transmittal of one or more communications to each anomalous agent; and store, in the federated database, a record set that includes a representation of the validator AI model subset, the series of computer-executable instructions, the anomalous agent set, and a timestamp of the identification of the anomalous agent set.

8 . The non-transitory, computer-readable storage medium of claim 7 ,
wherein the federated database is a federated ledger that represents multiple independent entities configured to control a hash-chained log, and wherein the federated ledger is configured to replicate a representation of the record set to a respective computing device associated with each of the multiple independent entities in response to a quorum co-signature from the multiple independent entities.

9 . The non-transitory, computer-readable storage medium of claim 7 , wherein the instructions further cause the system to:
establish a predefined schedule to rotate the validator AI model subset,
wherein the predefined schedule is established using one or more of: (1) time intervals or (2) a number of AI-based agents evaluated; and
using the predefined schedule, determine the validator AI model subset.

10 . The non-transitory, computer-readable storage medium of claim 7 , wherein the system is further caused to:
dynamically select the validator AI model subset using dimension-specific features of a particular dimension being evaluated.

11 . The non-transitory, computer-readable storage medium of claim 7 , wherein the system is further caused to:
obtain, for each AI-based agent, additional operational data stored in an external database different from the federated database,
wherein the additional operational data includes one or more of: a legal proceeding associated with a particular AI-based agent or a provider agent associated with the particular AI-based agent, and
wherein the validator AI model subset is configured to identify the anomalous agent set by comparing, for each AI-based agent, the additional operational data against the one or more allocation thresholds defined in the expected resource allocation data structure.

12 . The non-transitory, computer-readable storage medium of claim 7 , wherein the federated database is a federated le

Assignees

• Citibank Na

Inventors

• Mysore Vishal
• Rao Prithvi Narayana
• Jain Payal
• Uzzell Sawyer
• De Castro Marchese Joao Paulo
• Myers James

Classifications

• H04L67/1097

  for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS] · CPC title

• H04L47/125

  by balancing the load, e.g. traffic engineering · CPC title

• H04L47/122

  by diverting traffic away from congested entities · CPC title

• H04L47/70

  Admission control; Resource allocation · CPC title

• H04L67/10

  in which an application is distributed across nodes in the network (software deployment G06F8/60; multiprogramming arrangements G06F9/46) · CPC title