Panoptes
A watchdog that stops sensitive data from reaching AI assistants.
- Python
- Security
Problem
Employees and developers paste PHI, PII, financial data, and credentials into AI assistants like ChatGPT, Claude, and Copilot, creating HIPAA, PCI-DSS, and GDPR exposures before anyone can intervene.
Approach
Real-time monitoring of AI interactions across 25+ services and 10 browsers and IDEs, with 70+ detection patterns, ML confidence scoring to cut false positives, response scanning, a VSCode prompt-injection scanner, and optional blocking, wired into enterprise DLP and SIEM.
Outcome
v5.6 ships 70+ detection patterns and 572+ passing tests, with CEF/LEEF integrations for Splunk, Sentinel, and QRadar, Purview and Forcepoint DLP, centralized policy management, and a prompt-injection file scanner.
See it in action
Panoptes blocking a PII and PHI submission in real time. Nothing crosses the boundary.
Your team uses ChatGPT, Claude, and Copilot every day. The risk is what they paste in: a patient record, a customer's card number, an API key, a slab of production config. Once it is in the prompt, it is gone. Panoptes watches AI interactions in real time and catches sensitive data before it leaves the machine.
What Panoptes does
- Detects 70+ patterns across PHI, financial data, PII, and credentials, including international and industry-specific formats.
- Watches 25+ AI services across 10+ browsers and IDEs, scanning both prompts and the AI's responses.
- ML confidence scoring cuts false positives so alerts stay actionable.
- Optional blocking stops a risky request outright, and fails closed: a scan timeout blocks rather than allows.
- Scans workspace files for prompt-injection payloads before they reach a model.
- Central policy: define detection rules once on the server, enforce them on every client.
Works with your stack
Alerts route where your team already works: email, Slack and Teams webhooks, your SIEM (Splunk, Sentinel, QRadar, Elastic, Chronicle via CEF/LEEF), and enterprise DLP (Microsoft Purview, Forcepoint, Symantec, Digital Guardian).
Built to trust
Detection runs locally. Sensitive data never leaves the machine except through the alert channels you configure, and it is masked in logs. The code is open source and auditable. v5.6 ships behind 572+ passing tests, with coverage mapped to HIPAA, PCI-DSS, and GDPR.
Roadmap
Real-time watchdog detecting PHI, PII, financial data, and credentials in AI assistant interactions across browsers and IDEs.
Full roadmap →