Shadow AI discovery
Every AI tool reached from your network or through a corporate identity, including the ones nobody told you about.
Detect and Respond
AI Guardian
Shows you which AI tools your staff use, what they send to them, and gives you the controls to do something about it.
- Audience
- Security and governance
- Headline
- AI tools discovered
- Sources
- Proxy, cloud, identity, code
- Output
- Findings and policy
What it does
Inside AI Guardian
Discover and govern AI usage across the organization: detection, inspection, and enforcement.
Prompt inspection
What actually left in a prompt. Sensitive data, credentials, and source code detected in content rather than guessed from a domain.
Threat chains
Related AI events linked into one narrative, so a slow exfiltration across many sessions reads as a single story.
Prompt injection detection
Injection attempts against your own AI-facing applications, mapped to MITRE ATLAS techniques.
Per-user risk
Behavioural baselines per user, so an unusual volume or an unusual destination stands out without blanket blocking.
Policy and enforcement
Allow, warn, or block per tool, per group, per data class. Applied through the controls you already run.
How it works
See, decide, act.
Every agent follows the same three movements. What differs is what it reads, what it scores, and what waits for a person.
How it works
See, decide, act.
See
Every AI destination reached
Forward-proxy traffic, cloud AI service calls, identity events, and code-assistant usage, read together.
- Netskope and Zscaler web transactions
- Azure OpenAI and OpenAI usage
- GitHub Copilot activity
- Entra ID app consents
Decide
Content inspected, behaviour scored
Prompts and uploads are inspected for sensitive content, checked for injection, and weighed against that user's own baseline.
- Personal data detected in a prompt
- Credentials or keys in a paste
- Source code leaving for an unapproved model
- Volume far outside a user's normal
Act
Governed, not just observed
Findings open with a severity and a framework mapping. Policy decides whether a destination is allowed, warned on, or blocked.
- Block an unapproved destination for a group
- Warn the user in the moment
- Open a finding for review
- Map to ISO 42001 and the OWASP LLM Top 10
1 / 3
Exfiltration Map
Governance
How AI Guardian is governed
What always waits for a person differs meaningfully between agents, so it is stated per agent rather than as a platform-wide generality.
What it does on its own
Discovers destinations, inspects content, scores behaviour, opens findings, and applies policies you have already approved.
What always waits for a person
Creating or widening a block policy, and any enforcement affecting a whole group, requires an administrator.
What it records
Every detection with the engine that produced it, its framework mapping, the policy applied, and who changed that policy.
Works alongside
All of Detect and Respond →Alert Navigator
Detect and Respond · v2.0
Alert triage that separates true from false positives, groups related alerts into cases, and opens the ticket.
- Pending triage
- True positives
- Tickets opened
Phishing Remediation
Detect and Respond · v2.0
Multi-source phishing triage with enrichment, mailbox scoping, and analyst-gated remediation.
- Awaiting decision
- Incidents handled
- Actions reversed