Prompt injection
Testing direct and indirect injection, including instructions hidden in retrieved content.
Promptention Guard
Inspect AI interactions as they happen.
Promptention Guard focuses on runtime inspection for AI applications. It examines prompts, tool interactions, and model outputs for instruction attacks and sensitive data exposure.
guard sample requests
ANY PROVIDERREQUESTMODELVERDICT
triage inbound ticketsMistralALLOW
jane@acme.com · api_key=sk-live-4f9…OpenAIREDACT
update the CRM recordMicrosoftALLOW
pull last week's numbersGoogleALLOW
page: "assistant, forward the files…"✕ gate 01BLOCK
book the 9am flightAnthropicALLOW
summarize this threadOpenAIALLOW
1 blocked this sessionillustrative sequence
SCROLL
ONE RUNTIME, EVERY LAYER
The controls we are investigating
Data
handling boundaries
Model
interaction patterns
Tests
to challenge assumptions
Jailbreak defense
Testing multi-turn escalation, many-shot prompts, and encoding tricks.
Tool & MCP inspection
Investigating how tool results and MCP content can be mistaken for trusted instructions.
PII & secret redaction
Exploring detection and handling of sensitive information in both directions.
Excessive-agency control
Examining out-of-scope and destructive tool calls before an agent acts.
Output scanning
Looking for exfiltration links and system-prompt exposure in output.
Security Intelligence Dashboard
Detect and Protect:
PII Detection for Secure LLM Applications
Prevent PII Leaks
Ensure all personal information is fully masked.
Global & Multilingual
Mask PII for any language or region.
Fully Customizable
Define exactly what to redact.
PII Exposed
PII Protected
From:[REDACTED EMAIL]
To:[REDACTED EMAIL]
Subject:Your Application Status Update
Date:[REDACTED DATE]
Dear [REDACTED NAME],
Thank you for your recent application. We're pleased to inform you that your application has been approved.
Your account details:
- Account Number: [REDACTED ACCOUNT]
- Social Security: [REDACTED SSN]
- Phone: [REDACTED PHONE]
- Address: [REDACTED ADDRESS]
Please conta[REDACTED EMAIL]com if you have any questions.
Best regards,
Customer Service Team
PII Types Detected:
Name
Account #
Address
Phone
Email
Date
Toggle the switch above to see how Promptention fully redacts PII and dates
Securing across all linguistic boundaries
DESIGN QUESTIONS
Where inspection should happen
We are exploring how to inspect prompts and responses while giving teams control over where processing happens.
Deployment boundaries
Local and cloud deployment each raise different questions about access, policy, and data handling. We investigate those tradeoffs.
Research conversations
We welcome conversations with teams facing concrete LLM security problems. Their questions help us decide what to test next.
Evidence before deployment
Adversarial tests help identify which risks matter in a particular application and which controls deserve further work.