SERVICES / LLM SECURITY

LLM & AI security

We test how an AI feature can be prompted, what it can reach and what it can be induced to do—focusing on the security of the product, not model scores.

WHAT WE TEST

·Direct and indirect prompt injection

·Tool and function abuse

·Retrieval that ignores your permission model

·System prompt and customer-data leakage

·ML supply chain and model registry

BOOKED AS

Extension to Probe or Pipeline

3–5 days

OWASP LLM Top 10 · MITRE ATLAS · findings reviewed by an engineer.

Discuss this assessment

EVERY ENGAGEMENT

Signed authorization before technical work · public standards (PTES, OWASP, NIST, ATT&CK) · report reviewed by a second engineer · retest within 30 days included · engagement data destroyed 30 days after closure · delivery date and 10% late-delivery discount stated in the contract.