Before
A user or team member would likely need to review each prompt for PII, identify sensitive passages, and redact them before submitting the request for AI processing. They would also need to notify the requester about the redaction and manually copy the response back into the Slack conversation.
How it works
- Receive the Slackbot prompt and analyze its text for personally identifiable information.
- If PII is detected, replace the identified text with asterisks and notify the user; otherwise, continue with the original prompt.
- Submit the appropriate prompt for AI completion and post the generated answer to the originating Slack channel.
Workflow details
- Trigger and inspection: The workflow starts when a user submits a text prompt through the Slackbot `/gpt` command. It analyzes the prompt for personally identifiable information and uses the originating channel to return status messages and results.
- PII decision: If no PII is detected, the original prompt proceeds directly for AI processing. If PII is found, the workflow branches into a protection path before any AI request is made.
- Anonymization: For each detected entity, the workflow identifies the corresponding text segment and replaces the sensitive content with asterisks. The resulting prompt is treated as the version sent for AI processing.
- User notification: When anonymization occurs, the workflow posts a warning to the Slack channel, includes a copy of the prompt with detected PII obfuscated, and explains that the AI response will follow.
- AI response delivery: The protected or original prompt is submitted for completion, and the generated answer is posted back to the same Slack channel. This enables AI assistance while keeping detected PII out of the prompt sent for processing.
What changed
PII detection, branching, redaction, user notification, and response delivery now happen within the Slackbot flow before and after AI processing. The human requester receives either a direct answer or an explanation and obfuscated prompt without manually performing the privacy check or redaction.
Work removed
Manually reviewing each prompt for PII, locating and redacting sensitive text, notifying the requester about detected PII, copying the protected prompt for AI processing, and reposting the generated answer in Slack.
Expected outcome
Each request removes at least three recurring manual actions—PII review, sensitive-text redaction, and notification—while ensuring that prompts with detected PII are anonymized before AI processing.
