Deriving input restrictions for artificial intelligence agents
The invention provides a rules engine that manages user requests within an interface integrating multiple AI platforms and AI Models. Upon receiving a query, the engine assigns scores based on factors like management rules, user preferences, and contextual information. Determinative scores, such as those enforcing strict enterprise policies, can override others, leading the engine to block or reroute the query. If no score is determinative, the engine forwards the query and associated prompts to a specialized routing AI Model for contextual analysis. Based on this analysis, the rules engine directs the query to the most appropriate AI Model or defaults to the user-designated AI Model. This system balances user intent with rule enforcement, optimizing query processing across various AI platforms while ensuring compliance with enterprise guidelines.
1 . A method for controlling client-application-initiated execution of artificial intelligence (“AI”) agents, comprising:
receiving a request to generate a derived AI agent input restriction from a file that comprises an example restricted input, the generation request being received from an administrative user of an AI agent management service, and the file being received through an upload to a first user interface (“UI”) provided by the AI agent management service;
causing a derived restricted input marker to be identified based on an evaluation of the example restricted input of the file;
generating the derived AI agent input restriction by causing the AI agent management service to associate the derived restricted input marker with a management profile;
identifying a request from a client application to initiate execution of an AI agent, the request comprising a received input from the client application, wherein the received input originates from an end user through a second UI generated by the client application;
generating a confidence score based on a cumulative evaluation of multiple derived restricted input markers and the received input; and
enforcing the derived AI agent input restriction on the received input to the AI agent, wherein the enforcement comprises at least one of blocking and modifying the received input prior to the received input reaching an AI model, wherein the AI model is identified as part of executing the AI agent, and wherein the enforcing occurs based on determining that the confidence score exceeds a minimum confidence threshold for enforcement.
2 . The method of claim 1 , wherein enforcement comprises selecting a different AI model based on the confidence score, wherein different AI models have different minimum confidence thresholds.
3 . The method of claim 1 , wherein enforcement comprises selecting a different AI model based on the confidence score, wherein the request is blocked in an instance in which the confidence score does not satisfy the minimum confidence threshold for any of the different AI models.
4 . The method of claim 1 , wherein enforcement comprises selecting a different AI model based on the confidence score.
5 . The method of claim 1 , wherein the minimum confidence threshold is determined based on a group affiliation of a user of the client application.
6 . The method of claim 1 , wherein the AI model is an enterprise AI model, and wherein the enterprise AI model has a different minimum confidence threshold than a non-enterprise AI model.
7 . The method of claim 1 , wherein the confidence score is generated by a second AI model.
8 . The method of claim 1 , further comprising changing the minimum confidence threshold based on whether a user device is executing the client application within a geofence.
9 . The method of claim 1 , wherein the derived restricted input marker comprises content formatting, and wherein the enforcement occurs in an instance in which the received input satisfies a threshold similarity to the content formatting.
10 . The method of claim 1 , wherein the derived restricted input marker comprises a content type that comprises at least one of personally identifiable information (“PII”) and confidential information, and wherein the enforcement occurs in an instance in which the input to the AI agent satisfies a threshold similarity to the content type.
11 . The method of claim 1 , further comprising causing the second UI to display an option to override the enforcement.
12 . The method of claim 11 , wherein causing the UI to display the option is based on an evaluation of a user profile and an execution context associated with the client application.
13 . The method of claim 1 , wherein the derived restricted input marker comprises a content subject, and wherein the enforcement occurs in an instance in which the received input satisfies a threshold similarity to the content subject.
14 . The method of claim 1 , further comprising modifying the blocked input to remove content that triggered the enforcement, wherein the modifying occurs prior to sending the blocked input to the AI model.
15 . A non-transitory, computer-readable medium containing instructions that, when executed by a hardware-based processor, causes the processor to perform stages for controlling client-application-initiated execution of artificial intelligence (“AI”) agents, the stages comprising:
receiving a request to generate a derived AI agent input restriction from a file that comprises an example restricted input, the generation request being received from an administrative user of an AI agent management service, and the file being received through an upload to a first user interface (“UI”) provided by the AI agent management service;
causing a derived restricted input marker to be identified based on an evaluation of the example restricted input of the file;
generating the derived AI agent input restriction by causing the AI agent management service to associate the derived restricted input marker with a management profile;
identifying a request from a client application to initiate execution of an AI agent, the request comprising a received input from the client application, wherein the received input originates from an end user through a second UI generated by the client application;
enforcing the derived AI agent input restriction on the received input to the AI agent, wherein the enforcement comprises blocking the received input prior to the received input reaching an AI model, wherein the AI model is identified as part of executing the AI agent;
determining to include an option to override the enforcement based on at least one of a user profile associated with the end user or an execution context associated with the client application; and
causing the option to be displayed by the client application.
16 . The non-transitory, computer-readable medium of claim 15 , wherein blocking the received input comprises causing a notification to be transmitted to the client application, the notification describing the enforcement of the derived AI agent input restriction.
17 . The non-transitory, computer-readable medium of claim 16 , wherein the notification comprises the option to override the enforcement, the stages further comprising:
receiving a request to override the enforcement;
causing a logging of the request and the received input in association with an identifier of the end user; and
sending at least part of the blocked input to the AI model.
18 . The non-transitory, computer-readable medium of claim 15 , the stages further comprising modifying the blocked input to remove content that triggered the enforcement, wherein the modifying occurs prior to sending the blocked input to the AI model.
19 . The non-transitory, computer-readable medium of claim 15 , wherein the derived restricted input marker comprises a content subject, and wherein the enforcement occurs in an instance in which the received input satisfies a threshold similarity to the content subject.
20 . A system for controlling client-application-initiated execution of artificial intelligence (“AI”) agents, comprising:
a memory storage comprising a non-transitory, computer-readable medium comprising instructions; and
at least one hardware-based processor that executes the instructions to carry out stages comprising:
receiving a request to generate a derived AI agent input restriction from a file that comprises an example restricted input, the generation request being received from an administrative user of an AI agent management service, and the file being received through an upload to a first user interface (“UI”) provided by the AI agent management service;
causing a derived restricted input marker to be identified based on an evaluation of the example restricted input of the file;
generating the derived AI agent input restriction by causing the AI agent management service to associate the derived restricted input marker with a management profile;
identifying a request from a client application to initiate execution of an AI agent, the request comprising a received input from the client application, wherein the received input originates from an end user through a second UI generated by the client application;
generating a confidence score based on a cumulative evaluation of multiple derived restricted input markers and the received input; and
enforcing the derived AI agent input restriction on the received input to the AI agent, wherein the enforcement comprises at least one of blocking and modifying the received input prior to the received input reaching an AI model, wherein the AI model is identified as part of executing the AI agent, and wherein the enforcing occurs based on determining that the confidence score exceeds a minimum confidence threshold for enforcement.