Secure tunnel access to remote client resources for artificial intelligence agents

A non-transitory, computer-readable medium containing instructions for secure cloud connections that provide artificial intelligence (“AI”) agents with access to client resources, wherein the instructions are executed by at least one processor to perform stages comprising:

• receiving a user interface (“UI”) selection to create a managed cloud connector object in an AI platform, wherein the managed cloud connector object is assigned a key, wherein the UI presents an option to download a cloud connector based on the managed cloud connector object;
• receiving, from a cloud connector executing on a remote device, an authentication request at a connection manager that executes on a connection server, wherein the connection server is instantiated by an AI platform;
• opening a first secure tunnel between the cloud connector and the connection manager, wherein the cloud connector and connection manager encrypt traffic in both directions based on the key, and wherein opening the first secure tunnel requires opening no firewall ports at the remote device;
• receiving an input that triggers execution of an AI agent, wherein the AI agent is configured to access an AI model that executes at the remote device;
• evaluating the input with model guardrails, wherein the model guardrails are configured to detect prompt injection and personally identifiable information (“PII”);
• transmitting a modified input over the first secure tunnel from the connection manager to the cloud connector, wherein the modified input is based on the evaluation, and wherein the modified input comprises an input to the AI model;
• receiving, over the first secure tunnel, an output from the cloud connector, wherein the output originates from the AI model;
• evaluating the output with the model guardrails;
• and
• transmitting a result to the remote device for display, wherein the result is based on the output.

A system for secure cloud connections that provide artificial intelligence (“AI”) agents with access to client resources, the system comprising:

• at least one physical non-transitory, computer-readable medium comprising instructions;
• and
• at least one processor that executes the instructions to perform stages comprising:
• receiving a user interface (“UI”) selection to create a managed cloud connector object in an AI platform, wherein the managed cloud connector object is assigned a key, wherein the UI presents an option to download a cloud connector based on the managed cloud connector object;
• receiving, from a cloud connector executing on a remote device, an authentication request at a connection manager that executes on a connection server, wherein the connection server is instantiated by an AI platform;
• opening a first secure tunnel between the cloud connector and the connection manager, wherein the cloud connector and connection manager encrypt traffic in both directions based on the key, and wherein opening the first secure tunnel requires opening no firewall ports at the remote device;
• receiving an input that triggers execution of an AI agent, wherein the AI agent is configured to access an AI model that executes at the remote device;
• evaluating the input with model guardrails, wherein the model guardrails are configured to detect prompt injection and personally identifiable information (“PII”);
• transmitting a modified input over the first secure tunnel from the connection manager to the cloud connector, wherein the modified input is based on the evaluation, and wherein the modified input comprises an input to the AI model;
• receiving, over the first secure tunnel, an output from the cloud connector, wherein the output originates from the AI model;
• evaluating the output with the model guardrails;
• and
• transmitting a result to the remote device for display, wherein the result is based on the output.