The AI Application Security Program structures reviews of AI-enabled applications, model APIs, agents, retrieval-augmented generation systems, tools, data flows, and surrounding controls. It separates observed model behavior from application, authorization, integration, and data-handling failures.
Standards and scope
The configured system library uses OWASP AI Security Verification Standard requirements. AISVS provides a framework for verifying technical security controls in AI systems, while project scope records the actual model, application, interfaces, trust boundaries, and assessment target.
- AI application scope should identify models, providers, agents, tools, retrieval sources, APIs, users, privileges, and data boundaries.
- Requirements can address input and output handling, model behavior, data protection, access, deployment, monitoring, supply chain, and application integration.
- Garak probe results are candidates and do not independently prove exploitability or business impact.
- Prompt, response, model, and tool evidence should be minimized and scrubbed of secrets and unnecessary personal data.
- Manual reasoning, reproducibility, authorization context, and impact analysis remain essential.
Who this is for
- AI-security students, researchers, educators, red teams, and portfolio builders.
- Model, application, platform, data, and product-security engineers.
- Risk, governance, privacy, compliance, and responsible-AI professionals.
- Agencies, schools, public bodies, and regulated organizations deploying or procuring AI-enabled systems.
What voiqq provides
- AISVS-aware project setup and canonical requirement mapping.
- Manual finding entry, spreadsheet import, and supported garak candidate normalization.
- Fields for target, probe, prompt, response, affected flow, evidence, impact, likelihood, remediation, and retest.
- Assignments, comments, evidence, lifecycle state, validation, and history.
- AI application security assessment snapshots built from reviewed findings.
A practical workflow
- Document authorization, system purpose, model and provider, data flows, tools, users, privileges, and exclusions.
- Choose applicable AISVS requirements and safe test procedures.
- Validate probe candidates for reproducibility, affected boundary, realistic impact, and accurate mapping.
- Assign remediation across model, application, data, identity, and integration owners.
- Retest safely and review residual risks, limitations, and report language.
Put the framework into practice
Use the program for education, AI red teaming, internal product assurance, consulting engagements, procurement review, or governance evidence while keeping technical findings and remediation decisions connected.