AI Governance Institute
All governance templates →How do we govern agentic coding assistants and AI developer tools?

Implementation Kit

AI Developer Tool Register and Data Boundary Checklist

Which AI coding tools are allowed, for whom, and under what conditions. An approved tool register, a data boundary review template, a developer access tier matrix, and a vendor policy-change log. Locally installed CLIs are the blind spot this kit is built around.

Who this is for: The platform, security, or engineering-governance owner setting policy for AI developer tools.

Download the kit (Markdown) ↓4 artifacts. Every table also copies as CSV.

1. Approved AI developer tool register

Spreadsheet

The allowlist, with the conditions attached to each tool and each developer tier.

Template

ToolApproved for tiersDeployment mode requiredData boundary review dateConditionsNext review
<tool>T1 / T2 / T3SaaS / private cloud / on-premYYYY-MM-DD<e.g. no prod creds, telemetry off, repo allowlist>YYYY-MM-DD

Worked example

ToolApproved for tiersDeployment modeBoundary reviewConditionsNext review
Cloud IDE assistantT1, T2SaaS (enterprise tenant)2026-08-10training opt-out on; no files matching secret patterns2027-02-10
Local coding CLIT1 onlySaaS, zero-retention endpoint2026-08-12not permitted on machines with prod credentials; context scoped to the working repo2026-11-12
Self-hosted model pluginT1, T2, T3on-prem2026-09-01none beyond standard2027-03-01

Acceptance criteria

  • Only tools on the register may be used; anything else is a policy exception.
  • Each tool has a deployment mode and conditions matched to the developer tiers allowed to use it.
  • Every tool has a data boundary review date and a next-review date.

2. Data boundary review template

Spreadsheet

What actually leaves the machine per request, and what the vendor does with it.

Template

QuestionFinding
What code or data is transmitted per request (open file, whole repo, diffs, terminal output)?
Is transmission scoped by config, and does the scope hold in practice?
Does opt-out operate at transmission or only at retention?
Is submitted code used for model training? Is opt-out the default?
Retention period for prompts and code context?
Where is data processed and stored?
Vendor breach notification SLA?
Telemetry and analytics: what is collected, can it be disabled?

Worked example

QuestionFinding
Transmitted per requestopen file + up to 20 nearby files + last terminal command
Scoped by config?yes, "workspace only"; verified with a proxy capture
Opt-out levelretention only; content is still transmitted for inference
Used for training?no on the enterprise tier; opt-out is the default there
Retention0 days on the zero-retention endpoint; 30 days otherwise
Processing locationUS; EU option not available
Breach SLA72 hours
Telemetryusage metrics; can be disabled via MDM policy

Acceptance criteria

  • The review is based on observed behaviour (a proxy capture), not only vendor documentation.
  • It distinguishes transmission-level from retention-level opt-out.
  • The completed review is retained as a vendor due diligence artifact.

3. Developer access tier matrix

Spreadsheet

Tier developers by what they can reach, and match tool strictness to the tier.

Template

TierAccess heldPermitted tool categoriesRequired deployment mode
T1 (highest sensitivity)production credentials, customer data, signing keyson-prem or private-cloud AI only; no third-party SaaS inference on those machineson-prem / private cloud
T2internal APIs, proprietary source, no prod credsenterprise-tenant SaaS with training opt-out and secret filteringSaaS (enterprise)
T3sandbox and public code onlyany approved toolany approved

Worked example

Applied to three developers.

DeveloperTierAccess heldTool decision
Platform SRET1production credentialsLocal coding CLI removed from their machines; on-prem model plugin only
Product engineerT2internal source, no prodCloud IDE assistant on the enterprise tenant; secret filtering enforced
Contractor on the public SDKT3sandbox + public codeAny approved tool

Acceptance criteria

  • Every developer is assigned a tier based on actual access, reviewed when access changes.
  • T1 machines do not run third-party SaaS inference tools.
  • The matrix is enforced by device policy where possible, not left to the honour system.

4. Vendor policy change monitoring log

Spreadsheet

A record of vendor data-handling policy changes and the re-approval they trigger.

Template

Date noticedToolChangeAffects our data boundary?ActionRe-approved on
YYYY-MM-DD<tool><what changed>Y / Nre-review / restrict / no actionYYYY-MM-DD

Worked example

Date noticedToolChangeAffects boundary?ActionRe-approved
2026-08-28Cloud IDE assistantdefault retention changed 0 to 30 days on standard tierYconfirm enterprise tenant unaffected; re-review2026-09-02
2026-09-10Local coding CLIadded optional agent mode with shell executionYrestrict: agent mode disabled by policy pending reviewpending

Acceptance criteria

  • There is an active way to notice vendor policy changes (change-watch, release notes feed), not only annual review.
  • Any change affecting the data boundary triggers re-approval before continued use in sensitive environments.
  • The log ties each change to an action and a re-approval date.

Governance controls this kit produces evidence for

Completing the artifacts above gives you a head start on the evidence requirements for these controls.

DGC-006
DGC-006

The tool register and conditions govern AI-generated code and its license and provenance implications.

PRC-014
PRC-014

Egress-based discovery and the register are the shadow-AI inventory for developer tools, including local CLIs.

AGT-019
AGT-019

The data boundary review is the AI tool and plugin supply-chain risk assessment for dev tools.

SEC-003
SEC-003

The access tier matrix keeps sensitive code and credentials out of third-party inference pipelines.

PRC-007
PRC-007

The policy change log is the vendor governance-change monitoring process for dev tools.

This kit backs one playbook. Read the full guidance for the reasoning behind each artifact.

Decide what to implement next

Assess your governance gaps, then create an action plan with owners and target dates. Build and export without an account; sign in when you want to save your plan.

Start the AI governance assessment →