Skip to content

F7-LAS™ — Fuller 7-Layer Agentic Security

CI DOI maturity

Current status: Established reference model with a prototype reference implementation. The repository is not a production agent platform, control plane, SOC product, SIEM/SOAR replacement, or production-ready implementation.

Whitepaper version 4.0 is the current publication and is distinct from repository release v4.0.0. Zenodo is the authoritative publication record for the whitepaper; the Git tag identifies the independently versioned repository release.

F7-LAS™ is Anthony Fuller's vendor-neutral, protocol-agnostic security model for designing, reviewing, and governing agentic AI systems. It separates security responsibilities across seven layers so that generative output is not treated as authority to act.

This is independent personal work. It is not affiliated with, endorsed by, or representative of Microsoft or any other employer.

The seven layers

Layer Control concern
1 System Prompt — soft-policy intent and role boundaries
2 Grounding / RAG — source and epistemic constraints
3 Planner / Controller — bounded planning and delegation
4 Tools & Integrations — proposed action surface and argument contracts
5 External Policy Engine — deterministic PDP/PEP guardrails
6 Sandboxed Execution — scope and blast-radius containment
7 Monitoring & Evaluation — correlated evidence and assurance

Software supply-chain security is a cross-cutting supplemental domain, Layer S; it is not an eighth F7-LAS layer.

What is in this repository now

The canonical workflow provides one deliberately constrained executable Layers 1–7 path. It does not make the other examples executable or production-ready. The behavioral scenario matrix exercises the canonical enforcement path; the separate golden-dataset runner validates scenario structure and does not prove the described allow/deny behavior. Non-canonical examples remain illustrative unless they are explicitly reclassified and tested.

Executable versus illustrative

Repository area Current classification
Validation scripts Executable structural validation
Canonical Python + OPA path Executable for one synthetic, offline, read-only action
Other OPA/PDP/PEP code Partial prototype
Planner, tools, sandbox, telemetry Illustrative prototypes
Other policy-engine examples Illustrative, non-canonical patterns
Other end-to-end workflows Not provided
Production integrations or actions Not provided

Nothing in this repository should be connected to production data, identities, cloud resources, security platforms, or remediation systems without independent engineering and security review.

Validate the current repository

The canonical code supports Python 3.10 or later. To reproduce the CI dependency environment, use Python 3.12.14 in an isolated environment and install the hash-locked file:

python -m venv .venv
source .venv/bin/activate
python -m pip install --require-hashes -r requirements-ci.lock
python scripts/validate-supply-chain.py
python scripts/validate-documentation.py
python scripts/validate-citation.py
python scripts/validate-control-traceability.py
python scripts/validate-prompts.py
python scripts/validate-policies.py
python scripts/allowlist-validator.py
python scripts/validate-settings.py config/settings.yaml
python scripts/validate-contracts.py
opa check --strict config/policies/canonical-workflow.rego
python -m pytest -q

These commands validate the current prototype and repository structure. They are not evidence of production readiness or full behavioral enforcement.

Start here

Version and maturity boundaries

Versions belong to individual artifacts:

Artifact Current version/status
Seven-layer model Established design baseline
Whitepaper v4.0, current publication; v3.0 preserved as an immutable historical artifact
Implementation guide Draft; bundled with 4.0.0 but not independently versioned
Control catalog v0.1 draft
Executable reference implementation Prototype included in repository release v4.0.0
Repository release v4.0.0, independently versioned from Whitepaper v4.0

Repository version numbers do not silently change the whitepaper, control catalog, or schema versions.

Licensing, trademark, and attribution

  • Documentation, diagrams, the model, and other authored content are licensed under CC BY 4.0.
  • Executable source code and policy/configuration examples are licensed under the MIT License.
  • LICENSE.md defines the file-level boundary for mixed directories.
  • F7-LAS™ is a trademark of Anthony L. Fuller. Copyright licenses do not grant trademark rights or imply endorsement.

Preferred citation metadata is in CITATION.cff. The current Whitepaper v4.0 DOI is 10.5281/zenodo.22867553. Historical Whitepaper v3.0 remains preserved unchanged in the repository.

Contributing

Read CONTRIBUTING.md before proposing a change. Changes to the fundamental seven-layer model require explicit author approval and must not be inferred from implementation work.

About

Fuller 7-Layer Agentic AI Security Framework (F7-LAS)

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

2 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages