SAIL Framework: A Practical Guide for AI Security

What it is

SAIL is a process framework from Pillar Security that attaches security work to seven stages of the AI lifecycle: Plan, Code/No Code, Build, Test, Deploy, Operate and Monitor. The page says it was shaped by conversations with AI and security leaders over the past year. A long list of reviewers (mostly CISOs and security leads) is credited.

How it works

The page gives each phase more scope than the phase name suggests:

• Plan (policy and safe experimentation): includes dedicated threat modeling for novel failures, initial data and model governance, and a vetting process for introducing new data or models. It also sets up secure experimentation environments.
• Code/No Code (asset discovery): the inventory covers models, datasets, no-code platforms and code components, built in-house or sourced externally. The page ties it to emerging AI regulatory requirements.
• Build (AI-SPM): maps and graphs AI assets and their interconnections to show attack surface. Controls are prioritized per asset using the protection requirements set in Plan, and residual risks are identified.
• Test (red teaming): covers simulated attacks, attempts to bypass safeguards, and validation of security configuration (hardening). Depth should match the protection requirements of the business process the AI supports.
• Deploy (runtime guardrails): guardrails and security configuration are active at release.
• Operate (safe execution): sandboxing and zero-trust isolate agents, with coding agents and MCP servers named as examples, from critical infrastructure and sensitive data.
• Monitor (activity tracing): telemetry feeds anomaly detection and audit evidence, and detections trigger automated containment or rollback. It also flags end-of-life conditions, which start structured decommissioning and archiving.

flowchart LR
    P[Plan: protection requirements] --> B[Build: prioritize controls]
    P --> T[Test: red team depth]
    B --> T --> D[Deploy] --> O[Operate] --> M[Monitor]
    M -->|detection| R[Containment or rollback]
    M -->|end of life| X[Decommission]

What to do

The page gives no commands or configuration. Its usable ideas are:

• Record a protection requirement per AI-backed business process in Plan, then reuse it to rank controls in Build and set red-team depth in Test.
• Include no-code AI platforms in the asset inventory, not only code and models.
• Run agents and MCP servers in sandboxes with zero-trust isolation.
• Wire monitoring to automated containment or rollback, and define an end-of-life trigger for decommissioning.

Caveats

The page names no tools, test procedures, metrics or control catalog. It says SAIL sits on top of existing standards but does not say which ones or how they map. It is a vendor page that ends with an offer for a SAIL analysis on the Pillar Security platform. The page shows "Last updated: 05.01.2026", and the date format is ambiguous.