Overview/Register/Testing & Validation
Testing & Validation
Attack the AI, prove what it does under adversarial conditions, and prove it again after repair.
Tab H Services in this area
| Ref | Service | What it does |
|---|---|---|
| TV-01 | LLM / AI Application Penetration Testing | Tests LLM applications and APIs for prompt injection, data leaks, jailbreaks, excessive agency and API abuse. Tests again after repairs. |
| TV-02 | AI Red Teaming | Simulates real attackers against AI systems, agents and the people and processes around them. |
| TV-03 | Guardrail & Jailbreak Testing | Tries to go around the guardrails of AI models and applications with known and custom jailbreaks. |
| TV-04 | AI Infrastructure Penetration Testing | Tests the cloud, containers, APIs and hosting that operate AI systems. |
| TV-05 | AI Exposure Snapshot Scan | Does a quick, mostly automatic technical scan to show AI exposure across the environment. |
| TV-06 | AI Model Assurance & Validation | Tests if AI models are reliable and robust, and operate as intended, including drift and adversarial robustness. |
| TV-07 | AI Runtime Defense Analysis | Monitors AI in production and attacks it with adversarial inputs to test detection, logs and response. |
In detail Testing & Validation
-
TV-01
LLM / AI Application Penetration Testing
Tests LLM applications and APIs for prompt injection, data leaks, jailbreaks, excessive agency and API abuse. Tests again after repairs.
Deliverables: Test report with findings · root-cause recommendations · retest letter
-
TV-02
AI Red Teaming
Simulates real attackers against AI systems, agents and the people and processes around them.
Deliverables: Attack story · findings · detection gaps · recommendations
-
TV-03
Guardrail & Jailbreak Testing
Tries to go around the guardrails of AI models and applications with known and custom jailbreaks.
Deliverables: Jailbreak test results · guardrail tuning list
-
TV-04
AI Infrastructure Penetration Testing
Tests the cloud, containers, APIs and hosting that operate AI systems.
Deliverables: Test report · remediation list
-
TV-05
AI Exposure Snapshot Scan
Does a quick, mostly automatic technical scan to show AI exposure across the environment.
Deliverables: Snapshot report · top risks
-
TV-06
AI Model Assurance & Validation
Tests if AI models are reliable and robust, and operate as intended, including drift and adversarial robustness.
Deliverables: Validation report · robustness measurements · monitoring recommendations
-
TV-07
AI Runtime Defense Analysis
Monitors AI in production and attacks it with adversarial inputs to test detection, logs and response.
Deliverables: Runtime findings · detection rules · hardening plan
We are not accepting new clients at present, so no briefing is running on Testing & Validation. The services in this area are described above.
Read the availability note