Recipe

A recipe is a reusable security configuration that defines which checks, actions, and policy behaviors should be applied to AI traffic. It lets teams standardize guardrail logic across routes and workloads so enforcement can be managed centrally instead of being rebuilt in each application.

Expanded Definition

A recipe is a reusable security configuration that specifies which checks, actions, and policy behaviors apply to AI traffic. In NHI and agentic AI environments, recipes help standardize enforcement so the same guardrail logic can be applied consistently across routes, workloads, or services without rebuilding policy in every application.

Definitions vary across vendors, because some platforms use recipe to mean a bundled policy template while others use it to describe an ordered control pipeline. The operational meaning is the same: a recipe translates governance intent into repeatable enforcement logic. That makes it adjacent to policy-as-code, but narrower in practice because it often governs request-time decisions such as prompt inspection, tool restrictions, content classification, and escalation handling. For broader governance context, teams often map recipes to the intent of the NIST Cybersecurity Framework 2.0, especially where control consistency and measurable enforcement are required.

The most common misapplication is treating a recipe as a static configuration file, which occurs when teams reuse it across routes without validating whether the underlying model, tool chain, or data sensitivity has changed.

Examples and Use Cases

Implementing recipes rigorously often introduces standardisation overhead, requiring organisations to weigh faster governance and consistent enforcement against reduced flexibility for product teams.

• A customer support route uses one recipe to block secrets from being echoed into model outputs, while a separate route allows summarisation but disables external tool calls.
• An internal copilot applies a recipe that checks prompts for regulated data, then routes suspicious requests to human review before any agent action is executed.
• A platform team publishes a baseline recipe for all AI traffic, then extends it by workload to reflect different tool permissions and data classification levels.
• A security team updates a recipe after reviewing the Ultimate Guide to NHIs, using the same guardrail pattern across API keys, service accounts, and agent runtimes.
• A regulated workflow applies a stricter recipe for high-risk transactions, pairing content checks with explicit approval gates and audit logging aligned to NIST Cybersecurity Framework 2.0 principles.

Recipes are especially useful when multiple AI systems share similar guardrail requirements but still need workload-specific tuning. They reduce drift by making the intended control pattern reusable, reviewable, and easier to test.

Why It Matters in NHI Security

Recipes matter because they turn abstract governance into repeatable control logic for AI traffic, which is where many NHI failures begin. If a recipe is missing, inconsistent, or overly permissive, the result is often uneven enforcement across tools, routes, and agents. That creates gaps in prompt handling, secret exposure prevention, and action approval. NHIMG research shows that 79% of organisations have experienced secrets leaks, with 77% of those incidents causing tangible damage, underscoring how quickly weak enforcement can become an incident. The same research also reports that only 5.7% of organisations have full visibility into their service accounts, which makes standardised recipes even more important when agent behaviour spans multiple identities and systems.

Good recipes support Zero Trust by ensuring every request is evaluated according to defined checks instead of assuming trust based on source, application, or workload. They also help security teams prove that policy has been applied consistently across environments, rather than only documented in design reviews. The practical value becomes clear after a leak, abuse event, or policy bypass, when organisations discover that control logic was inconsistent across the very routes that mattered most. Organisations typically encounter uncontrolled AI traffic only after a secret leak or unsafe agent action, at which point recipe management becomes operationally unavoidable to address.

Related resources from NHI Mgmt Group

• Recipe-level remediation