Granted Patent B2

US 12,596,813 · App. 19/283,194 · Granted Apr 7, 2026

Inventors: Manjit Rajaretnam (Irving, TX); Sofia Rahman (New York, NY); William Cameron (Jacksonville, FL); James Myers (Clearwater, FL); Ryan Bergeron (New York, NY)

G06F21/577G06F21/552

Abstract

Claims (85)

1 . One or more non-transitory, computer-readable storage medium comprising instructions recorded thereon, wherein the instructions, when executed by at least one data processor of a system, cause the system to:

receive, from a plurality of autonomous computing agents deployed within an operational environment, a plurality of activity data streams corresponding to (i) actions and inactions performed by the plurality of autonomous agents and (ii) environmental stimuli captured by the plurality of autonomous agents using a plurality of sensors associated with the operational environment;

deploy a plurality of observing agents, the plurality of observing agents comprising at least one static observing agent configured to monitor for expected agent behavior and at least one dynamic observing agent configured to detect deviations from the expected agent behavior;

detect, in real time using at least one observing agent of the plurality of observing agents, a deviation of an autonomous agent of the plurality of autonomous agents from the expected agent behavior;

identify, via a root-cause analysis, a decision path responsible for the deviation, wherein the decision path is associated with the autonomous agent;

execute, using the identified the decision path, a mitigative action comprising at least one of interrupting execution of the autonomous agent, suspending the autonomous agent, shutting down the autonomous agent, or isolating a subsystem of the autonomous agent; and

execute a preventative action to block future decision paths similar to the identified decision path by the autonomous agent or by other autonomous agents of the plurality of autonomous agents, wherein the preventative action is based on the deviation, the root-cause analysis, and the mitigative action.

2 . The one or more non-transitory, computer-readable storage medium of claim 1 , wherein the instructions further cause the system to:

configure each observing agent of the plurality of observing agents with a respective observation modality selected from a plurality of observation modalities, wherein at least two observing agents of the plurality of observing agents are configured with different observation modalities; and

periodically perform, for the at least one dynamic observing agent of the plurality of observing agents, modifications to an observation parameter set comprising at least one of sampling frequency, data-collection granularity, vantage point within the operational environment, and obfuscation level, such that each modification is unpredictable to the plurality of autonomous agents.

3 . The one or more non-transitory, computer-readable storage medium of claim 1 , wherein the instructions further cause the system to:

cause a first observing agent of the plurality of observing agents to transmit a first observation summary directly to a second observing agent of the plurality of observing agents, wherein the first observing agent is a static observing agent and the second observing agent is a dynamic observing agent;

compare, at the second observing agent, the first observation summary generated by the first observing agent with a second observation summary generated by the second observing agent to detect a conflict between the first observation summary and the second observation summary, the conflict regarding at least one of the plurality of autonomous agents;

based on detecting the conflict, apply, at the second observing agent, a conflict-resolution rule set comprising a precedence rule and a tie-breaking rule to generate a resolved determination for the conflict; and

update an observation policy for the plurality of observing agents based on the resolved determination.

4 . The one or more non-transitory, computer-readable storage medium of claim 1 , wherein the instructions for executing the mitigative action further cause the system to:

generate a revocation signal that disables at least one credential, token, or network route previously assigned to the autonomous agent associated with the decision path;

reroute all subsequent communications originating from or directed to the autonomous agent; and

store the mitigative action, a timestamp, and the decision path, on an auditable ledger.

5 . The one or more non-transitory, computer-readable storage medium of claim 1 , wherein the instructions for executing the preventative action further cause the system to:

update a behavioral policy to include a rule that blocks the decision path and similar decision paths matching the decision path within a similarity threshold;

adjust a parameter of the autonomous agent to disincentivize future selection of the decision path or the similar decision paths; and

transmit the updated behavioral policy to the plurality of autonomous agents.

6 . The one or more non-transitory, computer-readable storage medium of claim 1 , wherein the instructions for detecting the deviation further cause the system to:

employ the at least one dynamic observing agent to perform anomaly detection on the plurality of activity data streams by comparing current activity metrics to a learned baseline to identify unexpected actions or inactions;

employ the at least one static observing agent to compare the actions and the inactions of the autonomous agent to a predefined set of expected behavior patterns; and

detect the deviation based on (i) the at least one dynamic observing agent detecting an unexpected action or an unexpected inaction by the autonomous agent and (ii) the at least one static observing agent failing to identify a corresponding expected behavior pattern from the predefined set of expected behavior patterns.

7 . A computer-implemented method for deploying a distributed observing platform that enables comprehensive monitoring and control of autonomous AI agent behaviors, the method comprising:

receiving, from a plurality of autonomous agents operating within an operational environment, activity data descriptive of agent behavior and environmental conditions;

deploying a plurality of observing agents for monitoring the plurality of autonomous agents, the plurality of observing agents comprising at least one static observing agent configured to monitor for expected agent behavior and at least one dynamic observing agent configured to detect deviations from the expected agent behavior;

detecting, by at least one of the plurality of observing agents, a deviation of an autonomous agent from an expected behavior pattern;

determining, based on an analysis of the activity data, a cause of the deviation;

performing a mitigative action with respect to the autonomous agent based on the cause of the deviation; and

performing a preventative action to block future deviations similar to the deviation by the autonomous agent or by other autonomous agents of the plurality of autonomous agents, wherein the preventative action is based on the deviation, the analysis, and the mitigative action.

8 . The method of claim 7 , further comprising:

configuring each observing agent of the plurality of observing agents with a respective observation modality selected from a plurality of observation modalities, wherein at least two observing agents of the plurality of observing agents are configured with different observation modalities; and

periodically performing, for the at least one dynamic observing agent of the plurality of observing agents, modifications to an observation parameter set comprising at least one of sampling frequency, data-collection granularity, vantage point within the operational environment, and obfuscation level, such that each modification is unpredictable to the plurality of autonomous agents.

9 . The method of claim 7 , further comprising:

causing a first observing agent of the plurality of observing agents to transmit a first observation summary directly to a second observing agent of the plurality of observing agents, wherein the first observing agent is a static observing agent and the second observing agent is a dynamic observing agent;

comparing, at the second observing agent, the first observation summary generated by the first observing agent with a second observation summary generated by the second observing agent to detect a conflict between the first observation summary and the second observation summary, the conflict regarding at least one of the plurality of autonomous agents;

based on detecting the conflict, applying, at the second observing agent, a conflict-resolution rule set comprising a precedence rule and a tie-breaking rule to generate a resolved determination for the conflict; and

updating an observation policy for the plurality of observing agents based on the resolved determination.

10 . The method of claim 7 , wherein performing the mitigative action further comprises:

generating a revocation signal that disables at least one credential, token, or network route previously assigned to the autonomous agent associated with the cause of the deviation;

rerouting all subsequent communications originating from or directed to the autonomous agent; and

storing the mitigative action, a timestamp, and the cause, on an auditable ledger.

11 . The method of claim 7 , wherein executing the preventative action further comprises:

updating a behavioral policy to include a rule that blocks the cause and similar causes matching the cause within a similarity threshold;

adjusting a parameter of the autonomous agent to disincentivize future selection of the cause or the similar causes; and

transmitting the updated behavioral policy to the plurality of autonomous agents.

12 . The method of claim 7 , wherein detecting the deviation further comprises:

employing the at least one dynamic observing agent to perform anomaly detection on the activity data by comparing current activity metrics to a learned baseline to identify unexpected actions or inactions;

employing the at least one static observing agent to compare actions and inactions of the autonomous agent to a predefined set of expected behavior patterns; and

detecting the deviation based on (i) the at least one dynamic observing agent detecting an unexpected action or an unexpected inaction by the autonomous agent and (ii) the at least one static observing agent failing to identify a corresponding expected behavior pattern from the predefined set of expected behavior patterns.

13 . The method of claim 7 , wherein the mitigative action comprises at least one of interrupting execution of the autonomous agent, suspending the autonomous agent, shutting down the autonomous agent, or isolating a subsystem of the autonomous agent.

14 . A system comprising:

a storage device; and

one or more processors communicatively coupled to the storage device storing instructions thereon, that cause the one or more processors to:

receive, from a plurality of autonomous agents operating within an operational environment, activity data descriptive of agent behavior and environmental conditions;

deploy a plurality of observing agents for monitoring the plurality of autonomous agents, the plurality of observing agents comprising at least one static observing agent configured to monitor for expected agent behavior and at least one dynamic observing agent configured to detect deviations from the expected agent behavior;

detect, by at least one of the plurality of observing agents, a deviation of an autonomous agent from an expected behavior pattern;

determine, based on an analysis of the activity data, a cause of the deviation;

perform a mitigative action with respect to the autonomous agent based on the cause of the deviation; and

perform a preventative action to block future deviations similar to the deviation by the autonomous agent or by other autonomous agents of the plurality of autonomous agents, wherein the preventative action is based on the deviation, the analysis, and the mitigative action.

15 . The system of claim 14 , wherein the instructions further cause the one or more processors to:

configure each observing agent of the plurality of observing agents with a respective observation modality selected from a plurality of observation modalities, wherein at least two observing agents of the plurality of observing agents are configured with different observation modalities; and

periodically perform, for the at least one dynamic observing agent of the plurality of observing agents, modifications to an observation parameter set comprising at least one of sampling frequency, data-collection granularity, vantage point within the operational environment, and obfuscation level, such that each modification is unpredictable to the plurality of autonomous agents.

16 . The system of claim 14 , wherein the instructions further cause the one or more processors to:

cause a first observing agent of the plurality of observing agents to transmit a first observation summary directly to a second observing agent of the plurality of observing agents, wherein the first observing agent is a static observing agent and the second observing agent is a dynamic observing agent;

compare, at the second observing agent, the first observation summary generated by the first observing agent with a second observation summary generated by the second observing agent to detect a conflict between the first observation summary and the second observation summary, the conflict regarding at least one of the plurality of autonomous agents;

based on detecting the conflict, apply, at the second observing agent, a conflict-resolution rule set comprising a precedence rule and a tie-breaking rule to generate a resolved determination for the conflict; and

update an observation policy for the plurality of observing agents based on the resolved determination.

17 . The system of claim 14 , wherein the instructions for performing the mitigative action further cause the one or more processors to:

generate a revocation signal that disables at least one credential, token, or network route previously assigned to the autonomous agent associated with the cause of the deviation;

reroute all subsequent communications originating from or directed to the autonomous agent; and

store the mitigative action, a timestamp, and the cause, on an auditable ledger.

18 . The system of claim 14 , wherein the instructions for executing the preventative action further cause the one or more processors to:

update a behavioral policy to include a rule that blocks the cause and similar causes matching the cause within a similarity threshold;

adjust a parameter of the autonomous agent to disincentivize future selection of the cause or the similar causes; and

transmit the updated behavioral policy to the plurality of autonomous agents.

19 . The system of claim 14 , wherein the instructions for detecting the deviation further cause the one or more processors to:

employ the at least one dynamic observing agent to perform anomaly detection on the activity data by comparing current activity metrics to a learned baseline to identify unexpected actions or inactions;

employ the at least one static observing agent to compare actions and inactions of the autonomous agent to a predefined set of expected behavior patterns; and

detect the deviation based on (i) the at least one dynamic observing agent detecting an unexpected action or an unexpected inaction by the autonomous agent and (ii) the at least one static observing agent failing to identify a corresponding expected behavior pattern from the predefined set of expected behavior patterns.

20 . The system of claim 14 , wherein the mitigative action comprises at least one of interrupting execution of the autonomous agent, suspending the autonomous agent, shutting down the autonomous agent, or isolating a subsystem of the autonomous agent.

Assignments (2)

Continuity (21)

References Cited (341)