Technical assessments

Assurance is a decision, backed by evidence.

SERVICES / 01

Focused engagements that reveal how AI systems behave, where they fail and what should happen before deployment.

What you receive

Findings that connect technical behaviour to operational decisions.

Every engagement is scoped to the system, its use case and its real attack surface. We provide clear evidence, prioritized risks and practical remediation guidance.

AI Security Assessment

Understand the security weaknesses across your complete AI application - not only the model.

We examine how models, prompts, data, APIs, tools, permissions, retrieval pipelines and application controls interact. The result is a prioritized view of exploitable weaknesses and the controls required to address them.

Threat modelling for the AI system and its integrations
Prompt injection and indirect prompt injection testing
RAG, API, tool and data-access security
Sensitive-data exposure and exfiltration pathways
Agent permissions, identity and excessive agency
Architecture, control and remediation review

AI Assurance Assessment

Determine whether an AI system behaves as intended under expected, unexpected and high-risk conditions.

We translate business, safety and operational requirements into testable evaluation criteria, then gather evidence about reliability, factuality, robustness, consistency and policy adherence.

Task-specific evaluation design
Reliability and behavioural consistency testing
Factuality and groundedness evaluation
Robustness under ambiguous or hostile inputs
Policy and control adherence testing
Deployment-readiness evidence and findings

AI Due Diligence

Independently examine the technical claims and risks behind an AI product, vendor or investment.

We help enterprise buyers, boards, investors and programme owners distinguish credible AI capability from unsupported claims and hidden system risk.

Product and architecture review
Evaluation-methodology review
Security and privacy control assessment
Evidence review for performance claims
Third-party model and platform dependencies
Decision-grade risk and evidence summary

Begin with evidence

Know what your AI system can do before an attacker does.

Start an assessment