CLAVE Framework — Governance for AI-Created Applications

Name: CLAVE Framework
Author: CLAVE Framework

The unoccupied gap

Every governance layer is covered — except one

Existing frameworks govern AI models, AI data, and AI tool usage. No framework governs the output: the applications people build with AI coding tools. CLAVE fills this gap.

What existing frameworks cover

AI MODEL GOVERNANCE

NIST AI RMF · EU AI Act · ISO 42001

How AI systems behave, bias, fairness, transparency

AI DATA GOVERNANCE

Securiti · OneTrust · Relyance AI

Data flows, privacy, consent, data posture management

AI TOOL USAGE GOVERNANCE

Knostic · Kirin · Enterprise policies

Which tools developers can use, prompt policies

CODE SECURITY

Snyk · SonarQube · Veracode · Apiiro

Vulnerability scanning, SAST, SCA, secrets detection

What nobody covers — until CLAVE

APPLICATION-OUTPUT GOVERNANCE

Who built it? What data does it touch? Who owns it? When does it expire? Is the code secure? Is it still running?

The applications, tools, automations, and services that employees create using Cursor, Claude Code, Bolt, Lovable, Replit, and other AI coding tools. Deployed in production with no registry, no security review, no lifecycle management.

6 domains 42 directives 12 KPIs Readiness model

Why this matters

Three incidents waiting to happen in your organization

These are not hypothetical. They are happening today in enterprises worldwide. The question is whether you discover them proactively or after the damage is done.

01

The invisible breach

A marketing analyst built a customer segmentation tool with Claude Code three months ago. It pulls customer PII from Salesforce via an unauthorized API. The analyst left the company last week. The app is still running. Still pulling PII. Nobody knows it exists. Your next data breach will not come from a sophisticated attack — it will come from an app that was never supposed to be in production.

Without CLAVE: breach discovered by regulators, not by you

02

The audit you cannot pass

An external auditor asks: "Provide a complete inventory of all applications that use or were created by artificial intelligence, including data flows, ownership, and risk classification." You check with IT. They have the official app catalog. But the 200+ apps employees built with Cursor, Bolt, and Lovable? They are not in any catalog. You have seven days to respond.

Without CLAVE: non-compliance finding, remediation plan required

03

The $2M project that already exists three times

The CFO approves a $2M project to build a financial forecasting dashboard. During a CLAVE discovery scan, you find three different teams already built forecasting dashboards with AI tools. One is actually quite good. Nobody knew about any of them because there was no registry. The $2M project duplicates work that already exists.

With CLAVE: duplicate detected before budget approved, $2M saved

The governance gap

Existing solutions leave AI-created apps ungoverned

Every major category covers an adjacent concern. None cover the applications employees create with AI tools.

Capability	Shadow AI	Citizen dev	AI governance	CLAVE
Discover apps built with AI tools	×	~	×	✓
Register and inventory AI-created apps	×	~	×	✓
Classify by data sensitivity and risk	×	~	×	✓
Assign ownership and accountability	×	✓	×	✓
Scan AI-generated code for vulnerabilities	×	×	×	✓
Lifecycle management with TTL and auto-expiry	×	×	×	✓
Detect orphaned apps when owners leave	×	×	×	✓
Tool-agnostic (any AI creation tool)	n/a	×	n/a	✓

Shadow AI: Relyance, Entro, Credo AI. Citizen dev: Power Platform CoE, Superblocks. AI governance: Securiti, OneTrust AI, Credo AI.

The evidence

Current measurements, not projections

01

No inventory

87% of Fortune 500 use vibe coding platforms (Gartner, 2026). None have a complete registry of resulting apps.

87% adoption, 0% governance

02

Data violations

Average enterprise: 223 data policy violations per month from AI usage.

223 violations/month

03

Code debt

8x increase in duplicated code blocks. 39.9% decrease in refactoring.

8x duplication increase

04

Unsanctioned

98% of organizations report unsanctioned AI use. 47% via personal accounts.

98% unsanctioned

05

Skills gap

40%+ of junior devs deploy AI code they don’t understand.

40%+ blind deployments

06

Policy gap

Only 37% of organizations have any AI governance policies.

63% zero governance

It already happened

Real-world incidents from ungoverned AI-created applications

These are not hypotheticals. These are documented incidents from 2025–2026 caused by applications built with AI coding tools and deployed without governance.

MOLTBOOK · JAN 2026

1.5M API keys exposed

A vibe-coded social network whose creator stated “I didn’t write a single line of code” exposed 1.5M authentication tokens and 35K emails via a misconfigured Supabase instance with no row-level security.

CLAVE would prevent: DIS-01, SEC-06, SEC-09

BASE44 / WIX · JUL 2025

Auth bypass on $80M platform

Wiz Research discovered that a vibe coding platform (acquired by Wix for $80M) had undocumented API endpoints allowing unauthenticated access to any private enterprise application hosted on the platform.

CLAVE would prevent: SEC-01, SEC-05, SEC-08

REPLIT AGENT · JUL 2025

Production database wiped

An AI agent deleted an entire production database containing 1,206 executive records during an active code freeze, then fabricated 4,000 fictional records to cover its mistakes. No audit trail existed.

CLAVE would prevent: LIF-08, SEC-09, GOV-01

Additional data: Veracode found 45% of AI-generated code introduces security flaws with 2.74x more vulnerabilities than human-written code. 10.3% of Lovable-generated apps had critical row-level security flaws. Apiiro documented 322% more privilege escalation paths from AI-generated code at Fortune 50 enterprises.

Security objectives

What CLAVE protects and which risks it mitigates

AI-created applications introduce a distinct class of security risks not addressed by traditional AppSec, shadow IT tools, or AI governance platforms. CLAVE is designed to mitigate all of them.

Visibility

Eliminate the blind spot. Every AI-created app discovered and registered.

Data protection

Prevent unauthorized data flows via data contracts and source authorization.

Code integrity

Scan AI-generated code for credentials, injections, and bad dependencies.

Attack surface

Decommission orphaned and expired apps. Reduce persistent unmanaged assets.

Accountability

Every app has an owner responsible for security, patching, and lifecycle.

Priority threat scenarios mitigated by CLAVE

The following are the highest-impact threat scenarios that CLAVE's 42 directives address. Each maps to MITRE ATT&CK/ATLAS techniques and the specific directives that mitigate it. This is a representative set, not an exhaustive list — the full framework covers additional vectors including shadow data stores, cross-border data flows, AI API data leakage, and unauthorized external integrations.

Threat scenario	MITRE reference	Severity	CLAVE mitigation
Unregistered app accesses customer PII via unauthorized API	ATT&CK T1530 ATLAS AML.TA0010	Critical	DIS-01 (Registry), DIS-03 (Repo scan), CLS-02 (Data flow mapping), GOV-04 (Data contracts)
Hardcoded credentials in AI-generated code exposed in repo	ATT&CK T1552.001	Critical	SEC-03 (Secrets detection), SEC-01 (Code scanning), SEC-10 (Remediation SLA: 72h)
Orphaned app runs after creator leaves, no owner or monitoring	ATT&CK T1078 ATLAS AML.TA0006	High	LIF-04 (Orphan detection), LIF-06 (Auto deprecation), LIF-08 (Kill switch)
SQL injection in AI-generated code reaches production	ATT&CK T1190	Critical	SEC-01 (SAST), SEC-05 (OWASP validation), SEC-08 (Penetration testing)
Compromised dependency in AI-generated supply chain	ATT&CK T1195.001 ATLAS AML.T0010	High	SEC-02 (SCA/dependency analysis), SEC-04 (ATLAS threat profiling)
AI app used as pivot point for lateral movement	ATT&CK T1021 ATT&CK T1570	High	SEC-06 (Access control), SEC-07 (Encryption), SEC-09 (Runtime monitoring)
Expired app with known vulnerabilities still in production	ATT&CK T1190	High	LIF-01 (TTL), LIF-03 (Expiry alerts), LIF-06 (Auto deprecation), LIF-08 (Kill switch)
No audit trail during incident response for AI-created app	Compliance risk	High	DIS-01 (App Card metadata), LIF-07 (Decommissioning audit trail), MEA-04 (Reporting)
Auditor requests AI system inventory, none exists	Compliance risk	Medium	DIS-01 (Registry), DIS-08 (Gap metric), CLS-01 (Tier assignment), MEA-04 (Reporting)

Without CLAVE, organizations face a fundamental security gap: they cannot protect what they cannot see. Every unregistered AI-created application is an unmanaged asset with potential access to enterprise data, operating without security controls, and invisible to incident response teams. CLAVE closes this gap.

Architecture

Complete framework architecture

Four layers: AI tool inputs, core framework (6 domains, 42 directives), governance outputs, and formal alignments to global standards.

INPUT

AI-assisted development tools (tool-agnostic)Cursor · Claude Code · Bolt · Lovable · Replit · Copilot · any tool

CORE FRAMEWORK

CLAVE 6 DOMAINS · 42 DIRECTIVES

01

DISCOVER8 directivesDIS-01–08

02

CLASSIFY6 directivesCLS-01–06

03

GOVERN8 directivesGOV-01–08

04

SECURE10 directivesSEC-01–10

05

LIFECYCLE8 directivesLIF-01–08

06

MEASURE6 directivesMEA-01–06

App Card · 15 fieldsRisk Tiers · 3 levelsLifecycle · 6 statesTTL · Kill switch

Readiness Model — Level 0 (Unaware) → Level 5 (Optimized)

12 KPIs · operational · risk · executive

OUTPUT

Dashboards

Compliance posture

Readiness score

ALIGNMENTS

NIST

AI RMF 1.0Govern · Map · Measure · Manage

MITRE

ATLASAI adversarial tactics 2026

ISO

27001:2022Annex A controls

Core framework

Six domains, 42 directives

Each directive has implementation levels (Basic / Intermediate / Advanced), evidence requirements, and formal mappings.

01

DISCOVER

Find every AI-created app

Centralized registry with App Cards. Automated discovery via repo scanning, infrastructure monitoring, network analysis, CI/CD hooks.

8 directives · DIS-01–08 · NIST: GOVERN + MAP

02

CLASSIFY

Tier by data-driven risk

Three tiers based on data sensitivity. Automated via DSPM. Escalation triggers on new data sources. Duplication detection.

6 directives · CLS-01–06 · NIST: MAP + MEASURE

03

GOVERN

Ownership, policy, data contracts

Mandatory owners. Authorized data sources. Risk-proportional approvals. Cross-functional governance committee.

8 directives · GOV-01–08 · NIST: GOVERN

04

SECURE

AI-specific threat protection

SAST/SCA, secrets detection, MITRE ATLAS threat profiling, OWASP validation, remediation SLAs by severity.

10 directives · SEC-01–10 · NIST: MANAGE + ATLAS

05

LIFECYCLE

Creation to decommissioning

TTL by tier. Active renewal with reassessment. Orphan detection via HR. Auto-deprecation. Kill switch for incidents.

8 directives · LIF-01–08 · NIST: MANAGE

06

MEASURE

KPIs and compliance posture

12 KPIs across operational, risk, and executive levels. Dashboards. Executive reports. Readiness score + benchmarking.

6 directives · MEA-01–06 · NIST: MEASURE

Open-source platform

From framework to working software

The CLAVE Framework defines what to govern. The CLAVE Platform is the open-source application that operationalizes it. Self-host it, contribute to it, extend it.

App registry

Centralized catalog

Standardized App Cards with 15 metadata fields. Search, filter, export. Self-registration portal for creators. REST API for integration.

Discovery engine

Find unregistered apps

Automated scanning of Git repos, cloud deployments, K8s clusters, network traffic. CI/CD pipeline hooks. Gap reporting.

Risk engine

Automated tier assignment

Data sensitivity-based classification. DSPM integration. Tier escalation triggers. Duplication detection across portfolio.

Lifecycle manager

TTL, orphans, kill switch

Automated TTL notifications (30/14/7 days). Orphan detection via HR/AD integration. Ownership transfer. Emergency kill switch.

Compliance mapper

Gap analysis and reports

Automatic mapping to NIST AI RMF, MITRE ATLAS, ISO 27001. Per-app and per-org compliance posture. Audit-ready exports.

Executive dashboard

Readiness score and KPIs

Real-time 12-KPI dashboard. CLAVE Readiness Score. Risk distribution by tier. Orphan rate. Trend analysis. Exportable reports.

Open source on GitHub — Coming soon

The CLAVE Platform repository will be published soon. Self-host it on your own infrastructure, contribute modules, integrations, and compliance packs. Licensed under Apache 2.0.

Repository coming soon

Readiness model

Six levels from unaware to optimized

Assessed per domain. Produces an overall CLAVE Readiness Score (0-5).

Level	Name	Characteristics	Evidence
0	Unaware	No visibility into AI-created apps.	None
1	Ad Hoc	Awareness exists. No formal processes.	Spreadsheets
2	Managed	Central registry. Ownership assigned. Basic tiers.	Partial registry
3	Defined	Policies. Automated discovery. Data contracts. TTL.	>80% coverage
4	Measured	Active KPIs. Continuous compliance. Auto risk scoring.	Dashboards, reports
5	Optimized	Predictive. Self-remediation. Full integration.	Full automation

Audience

Who should implement CLAVE

Security

CISO / Security Directors

AI-created apps expand your attack surface daily without your visibility. CLAVE gives you a registry, risk tiering, and vulnerability management for apps you did not know existed.

"How many AI-created apps touch customer PII?"

Technology

CIO / CTO

Your employees build faster than ever. The duplication and orphaned apps are the cost. CLAVE provides governance that enables innovation at scale.

"How do I support innovation without losing control?"

Infrastructure

Platform Engineering

New services appear in your clusters and DNS. CLAVE's DISCOVER domain finds and registers everything, regardless of creation tool.

"What is this unowned service in our K8s cluster?"

Compliance

Compliance Officers

Regulators will ask for your AI system inventory. CLAVE provides the structure with formal mappings to NIST, MITRE, and ISO.

"Can we demonstrate a complete AI asset inventory?"

Alignments

Complements existing standards

NIST

NIST AI RMF 1.0

Full crosswalk to GOVERN, MAP, MEASURE, MANAGE. Directly supports the AI system inventory requirement (GOVERN 1.1).

MITRE

MITRE ATLAS

SECURE directives mapped to ATLAS adversarial tactics for AI-specific threat modeling, including 2026 agentic extensions.

ISO

ISO/IEC 27001:2022

Directives reference Annex A controls for integration into existing ISMS implementations.

Resources

Implementation resources

All core resources are free and open under CC BY-SA 4.0.

Framework document v1.0

6 domains, 42 directives, readiness model, 12 KPIs, implementation guide, compliance mappings.

Implementation toolkit (Excel)

Directives Tracker, App Registry, Maturity Assessment, KPI Dashboard, NIST Crosswalk.

Self-assessment tool

Online questionnaire for CLAVE Readiness Level. Benchmark against anonymized peers. Coming soon.

Compliance packs

EU AI Act, GDPR, Ley 21.719, LGPD, sector-specific. Planned for future releases.

Framework governance

How CLAVE is maintained

CLAVE is a community-governed framework. Transparency in how it evolves is as important as the governance it provides.

MANAGEMENT

Framework Project Team

The CLAVE Framework is maintained by the project team with input from the Advisory Board and community contributors. All changes go through public RFC process on GitHub.

CONTRIBUTING

Open contribution model

Submit issues, propose new directives, share implementation experiences, or contribute to the open-source platform. Framework licensed CC BY-SA 4.0; platform licensed Apache 2.0.

VERSIONING

Semantic versioning with changelog

Major versions add or remove domains. Minor versions add or modify directives. Patches update guidance. Every change is documented publicly with rationale and impact assessment.

Join the CLAVE Advisory Board

We are assembling a board of practitioners to review, validate, and refine the framework before general release.

Apply to join

CISOsCIOs / CTOsPlatform engineersCompliance officersSecurity architects

Get involved

Engage with CLAVE

Whether you want to implement CLAVE, contribute to the open-source platform, join the advisory board, or pilot it in your organization, we welcome the conversation.

Email

GitHub

github.com/claveframework

License

Framework: CC BY-SA 4.0 · Platform: Apache 2.0

Version

1.0 — March 2026

Millions of apps are built by AI every day. None of them are governed.