Shadow AI discovery

Shadow AI is any AI tool used for work without IT or security approval: personal chatbot accounts, niche AI writing and transcription tools, AI browser extensions, local models, and AI features quietly switched on inside SaaS apps you already use. PromptGlass discovers them from every sensor and scores each one with a continuously updated catalog of AI apps.

Discovery sources

Source	Finds
Forward proxy and SWG chaining	Any AI destination, including API calls from scripts
Browser extension	AI web apps, personal vs corporate accounts, other AI browser extensions
Endpoint agent	Desktop AI apps, IDE assistants, local model runtimes and model files
SaaS connectors	AI features inside sanctioned platforms (Copilot, Gemini)
Imported logs	Firewall, DNS or SWG logs (CSV/syslog) for a quick retrospective

The AI app catalog

Each catalog entry includes:

• Identity: vendor, domains, desktop binaries, extension IDs, API hosts
• Data handling: whether inputs are used for training by default, opt-out options, retention, data-residency options, sub-processors
• Security posture: SSO/SCIM support, an enterprise tier, published security attestations, breach history
• Risk score: low, medium, high or critical, computed from the factors above

You can override any score and add internal notes. Unknown AI-like apps found by heuristics go to a review queue.

Sanctioning workflow

1. Discover: new apps appear under Discover with first-seen date, user count and departments.
2. Review: an owner reviews the catalog data, and a decision request can open a ServiceNow or Jira ticket.
3. Decide: mark the app sanctioned, tolerated or prohibited.
4. Enforce: policies use sanctioned: false or appRisk to coach or block. See policy examples.

Retrospective discovery

Before deploying any sensor, you can import 30 days of proxy or DNS logs to size the problem:

promptglass discover import --format zscaler-nss --file weblogs-oct.csv
promptglass discover report --since 30d --format html > shadow-ai-baseline.html