// SERVICE
AI Security Integration
Assessments for LLM apps and agents: prompt injection, tool abuse, data exfiltration.
// WHAT WE COVER
Coverage
01
Prompt injection
Direct and indirect, through documents, web pages and tools.
02
Tool permission review
What the agent can call, and what it should not.
03
Data exfiltration
Where private context leaks, and how.
04
Model supply chain
Weights, adapters, dependencies and their provenance.
05
Guardrail validation
Bypass attempts against filters and classifiers.
06
Red-team prompt corpus
Reusable, regionalised, updated each engagement.
// HOW IT RUNS
How an engagement runs
Assessments for LLM apps and agents: prompt injection, tool abuse, data exfiltration, model supply chain.
- Scoping call and rules of engagement.
- Testing with daily syncs.
- Executive and technical report.
- Free retest within 90 days.
DELIVERABLES
- OWASP Top 10 for LLM Applications
- Agent tool-permission review
- Red-team prompt corpus
- Guardrail validation
// COMMON QUESTIONS
FAQ
Do you test the model or the application?
The application and its integrations. Model-level testing is included when you host your own weights.
Can you assess an agent with tools?
Yes, that is most of the work now: tool abuse and permission scoping.
How are findings rated?
CVSS 3.1 where applicable, plus a business-impact narrative for the board.