Security Audit Records in GPTfy

Security audit records in GPTfy play a crucial role in tracking every AI interaction within Salesforce. Whenever a prompt is executed, and a response is received, a corresponding audit record is automatically created. These records help your team:

• Understand what data was sent and received
• Capture feedback and errors
• Monitor token usage and response time
• Maintain compliance and transparency

When Are Security Audit Records Created?

A security audit record is generated every time:

• A prompt runs, and the AI returns a response
• An error is encountered during prompt execution
• A user submits feedback on a response

These records ensure traceability of all prompt interactions, including data inputs, outputs, and user feedback.

Identifying Important Tabs and Fields in Security Audit Records

Understand the critical tabs and fields within these records to capture essential information.

Details Tab:

Field	Description
AI Prompt	The prompt text that triggered the record
Prompt Command	Command executed as part of the prompt
Message	Any error message from prompt execution
Record ID	Salesforce record linked to the prompt
GPT Config	AI connection used for the execution
Prompt Tokens	Number of tokens in the input
Completion Tokens	Tokens used in the AI response
Total Tokens	Combined token count (input + output)
Pre-Processing Time	Time to extract Salesforce data (in seconds)
Processing Time	Time taken by AI to generate the response
Post-Processing Time	Time to display/update data in Salesforce
Total Response Time	Sum of all processing times
Status	Final status (Processed, etc.)

Context Tab:

• Data (PII Key): The encryption key to process Personally Identifiable Information (PII).
• Data (Original): The raw data extracted from Salesforce.
• Data (PII Removed): Review the data processed to remove PII and transmit it to the AI.

The context tab has the option to View in the JSON Viewer. The Data Original and the Data PII Removed are available in JSON format, which allows users to view and analyze the JSON (JavaScript Object Notation) data in a user-friendly manner.

The viewer would show the raw JSON data on the left and a formatted, structured view of the data fields on the right.

Response Tab:

• Data (Decrypt PII Key): Decrypting key to reintroduce PII into the data.
• AI Processed Data (No PII): Examine the response after data deidentification.
• AI Processed Data (PII Added): Review the response received from the AI, now including PII.

The response from the Console will be populated at the top of the Response tab of the security audit record.

This will show a summary of the account, related objects, response log, and AI processed data with and without PII.

Prompt Actions:

When a prompt with actions is executed, corresponding security audit records are generated. These records document all details related to the specific action performed.

• Prompt Action Detail: This tab displays information about the action based on the selected prompt action.
• Completed: Indicates successful execution of the action.
• Errored: Indicates an issue during execution.
• Message: Displays error messages, if any, related to the execution of the action.
• Prompt Action Details: Shows the Prompt action details according to the prompt action selected.

The screenshot would display information about a prompt action, including type, status, message, and prompt action details.

Polling Tab:

The new Polling feature is designed to handle cases where the AI cannot initially process a user’s voice recording. This asynchronous process ensures multiple attempts to extract the transcript.

Feedbacks Tab:

Access a list of feedback related to a specific response. Feedback is captured through the thumbs-down icon on the GPTfy console, allowing users to provide suggestions if they have any. This feedback is associated with responses and linked to the respective audit record.

Feedback can be added via the GPTfy console, and here is the step-by-step procedure on how to provide feedback on a particular response. The feedback will be linked to the security audit record.

File Tab:

When a user sends a huge amount of data to AI, that is more than 131072 characters (character limit of a Long text area field), files will be created for those fields. Example:

• Data (Original)
• Data (PII Removed)
• AI Processed Data (No PII)
• AI Processed Data (PII Added)

Step-by-Step Process

1. Run a prompt
   Run a prompt on a record.
   The response gets pasted to the console, and a Security audit record is created.

   

2. Give Feedback

   • Click on the Thumbs-down icon in the console.
   • Add Comments and select any available options.
   • Click Save.

   

3. Verify the feedback

   • Open the security audit record by clicking on the hyperlink on the console.
   • Click on the feedback tab.
   • A related record of feedback is added.