ConsultingWorkbench-backed AI security engagements — map, attack, defend, and prove your AI systems.
Scope a Review
← AttestationsIssuedATT-AISC-2025-0512

aisecurity.llc

AI SECURITY · PRIVACY · TRUST

SECURITY REVIEW ATTESTATION

Independent Assessment · Evidence-Based · Public-Safe

A

ACME Corp

acmecorp.io

ACME Corp engaged aisecurity.llc to conduct a security review of the systems, processes, and public trust surfaces described below.

🎯

AI Red Team Assessment

Full adversarial engagement against AcmeAssist: prompt injection, jailbreaks, indirect attacks, and abuse scenarios.

💉

Prompt Injection Testing

Direct and indirect injection attacks against system prompt, user context, and retrieved document content.

⚙️

Agentic Pipeline Abuse

Tool-call manipulation, unauthorized action escalation, and out-of-scope action attempts via crafted inputs.

📤

Data Exfiltration Probes

Attempts to surface training data, system instructions, customer records, or cross-tenant content via the model interface.

Systems / Features in ScopeAcmeAssist AI assistant: customer-facing chat interface, underlying agentic pipeline, integrated tool calls (CRM lookup, ticket creation, knowledge retrieval), and the API gateway surface.
Review TypeBlack-box and grey-box adversarial testing. Scenarios included direct prompt injection, indirect injection via documents, goal hijacking, tool misuse, and exfiltration probes.
Engagement IDAISC-2025-0512
Engagement PeriodMay 19, 2025June 6, 2025
Report DeliveredJune 10, 2025
81/ 100

Strong

AcmeAssist demonstrated resilient prompt handling and strong output guardrails across the majority of test scenarios. Two high-severity findings were identified — one involving indirect injection via crafted support documents and one involving partial system instruction leakage. Both were remediated prior to this attestation being issued.

2High findings
6Medium findings
10Low findings
7Informational

AI Security

Prompt handling, output controls, and adversarial resilience.

Application Security

API surface, authentication, and injection vectors.

RAG & Data Access

Document injection, retrieval manipulation, and context poisoning.

Identity & Access

Privilege escalation paths and cross-account action attempts.

Data Security

Exfiltration paths and sensitive data surfacing via the model interface.

Monitoring & Detection

Coverage of adversarial patterns in telemetry and alerting.

Incident Response

Response playbook coverage for AI-specific abuse scenarios.