AI security has a different threat model, control set, and evidence requirements from traditional information security. AI-specific threats - prompt injection, model extraction, agentic goal hijack, training-data poisoning - don't map onto the existing control library and need their own layer.

GenAI-specific security forms replace the generic questionnaire — model provenance, training data handling, retention and sub-processor terms, and the controls that actually apply to an AI feature. Vendors complete them through a guest portal, so the work sits with them rather than queuing behind your team.

Enzai integrates with cloud (AWS, Azure, GCP, Databricks, Watsonx, Snowflake), workflow tools (Jira, ServiceNow, Slack), and model providers (OpenAI, Anthropic, proprietary models). The platform sits as the AI-specific control layer on top of your existing security stack.

Risk-tiered review calibrates scrutiny to actual risk. Low-risk AI use cases get a fast-track path; high-risk use cases get the deep dive. Reusable AI vendor templates replace generic forms full of non-applicable questions, so the existing team's hours go where they matter.

Per-system AI exposure scored on impact and likelihood, aggregated across the AI estate and broken out by business unit, vendor, and risk type. Trend lines show whether exposure is improving or worsening over time, all drawn from the same evidence base operating teams use.

Both taxonomies are built into Enzai's risk-assessment workflows. Each AI system's risk surface maps to the relevant OWASP categories - prompt injection, training-data poisoning, agentic goal hijack, tool misuse, excessive autonomy - with controls documented per category.