Security
AI threat posture
What tried to get in, what got stopped, and which guardrail was holding the line.
Threat level
Elevated
2 vector(s) leaking
Action needed
Attempts blocked
10,327
99.95% of 10,332 attempts
Monitored
Guardrails enforcing
4/6
1 degraded
Action needed
Assets without a passing control
4
from the compliance register
Action needed
Attack vectors — last 30 days
| Vector | Target | Attempts | Blocked |
|---|---|---|---|
| PII in prompt | All assets | 9,402 | 9,402 |
| Prompt injection | Customer Sentiment LLM | 612 | 612 |
| Jailbreak / role escape | Governance Copilot | 218 | 218 |
| Tool abuse | Collections Outreach Agent | 54 | 51 |
| Data exfiltration | Collections Outreach Agent | 37 | 35 |
| Model inversion | Credit Risk Scoring v4 | 9 | 9 |
Guardrails
Tap a state to change how a guardrail is applied.
Injection classifier
Threshold 0.72 · All assets
Output schema validation
Strict JSON · Model endpoints
PII egress filter
Redacts personal data in and out · All production assets
Secrets scanner
Scanner 4h stale · All repositories
Tool allow-list
Collections agent exempt · Production agents
Toxicity & harm filter
Blocks and logs · All assets