# GuardGPT

- **Event:** [Global Codex Hackathon: [San Francisco]](https://cerebralvalley.ai/e/openai-hackathon-sanfrancisco)
- **When:** Thu, Apr 16 at 10:00 AM – 8:00 PM (PDT)
- **Where:** 1515 3rd Street, San Francisco, CA
- **Team:** [Jordan Bredehoeft](https://cerebralvalley.ai/u/jordanbrede), [Peter Griffin](https://cerebralvalley.ai/u/cadet), [Sreevidhya Sreedharan](https://cerebralvalley.ai/u/sree_14), [Rob Koch](https://cerebralvalley.ai/u/robcube), [karthik mannava](https://cerebralvalley.ai/u/karthikmannava83)
- **Demo video:** https://youtu.be/eBORDPew3EQ
- **Gallery:** https://cerebralvalley.ai/e/openai-hackathon-sanfrancisco/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/openai-hackathon-sanfrancisco/hackathon/gallery/2

This project demonstrates how governed AI work can happen safely on sensitive data. It starts with a database's role-based access, where the same workflow shows masked or unmasked PII depending on the user’s role. Before anything reaches the model, copied prompts are inspected for hidden non-printing characters and malicious prompt-injection content. The system preserves the safe part of the prompt, removes the dangerous segment, and shows both the original and scrubbed versions for transparency. When data is retrieved from the database, dangerous stored text is also detected and quarantined before it reaches the AI summary path. OpenAI Guardrails protect the model boundary, while the application traces suspicious database objects, query history, and API activity to reveal where PII exposure may still occur outside the governed workflow. Together, those features turn governance from a passive control into an active remediation experience.

---

Markdown version of https://cerebralvalley.ai/e/openai-hackathon-sanfrancisco/hackathon/gallery/2. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt