Runnable now
7 canonical governance scenarios
Stable public conformance fixtures for deterministic allow, approval-required, policy-deny, and engine-guard outcomes.
Runnable public proof
Three proof surfaces. Three different questions.
The counts are intentionally separate. The 24 inference-integrity cases are not part of the 27 general demonstration scenarios.
Runnable now
9 packs / 27 scenarios
Broader synthetic governed-action coverage across documentation, source, CI/release, dependencies, secrets, tools, runtimes, administration, and adversarial declarations.
Runnable now
24 inference-integrity scenarios
Recorded provider-neutral evidence for identity, seed/token integrity, verifier trust, routing, supervisor/specialist localization, re-verification, operating modes, and recovery.
Reproduce everything: use the Governance Lab full-verification runbook for signed evaluator acquisition, expected counts, controlled mismatch checks, evidence generation, and checksums.
Responsibility boundaries
Policy, validation, and execution have distinct owners.
Keeping them separate prevents the demonstration harness or runtime from becoming a self-certifying policy authority.
Agent request
Anthesis governance boundary
Bounded execution
Evidence and outcome
Governance Lab independently tests the public Anthesis contract and is not in the runtime critical path.
| Component | Responsibility |
|---|---|
| Anthesis | Policy authority, deterministic evaluator semantics, approval requirements, capabilities, evidence semantics, and provenance. |
| Governance Lab | Independent conformance, synthetic scenario packs, inference-integrity fixtures, reports, and walkthroughs. It does not execute production effects. |
| Dubnium | Reference runtime, gateway, bounded tools, execution, and runtime evidence. It consumes decisions and does not define policy authority. |
Integration assurance
Where enforcement lives is not the same as how strong it is.
The same integration style may be advisory, moderate, strong, or runtime-enforced depending on which bypass paths remain.
| Mode | Enforcement location | Required bypass control | Typical assurance |
|---|---|---|---|
| Tool wrapper / invoke | Tool surface | Raw effectful tools are unavailable to the agent. | Moderate to strong |
| MCP mediation | Tool surface | Raw downstream MCP servers, credentials, and direct service paths are unavailable or constrained. | Moderate to strong |
| Gateway / sidecar | Infrastructure | Downstream effects are unreachable except through the governed gateway. | Strong |
| Capability tokens | Infrastructure / downstream tool | Effects reject missing, expired, altered, replayed, or out-of-scope grants. | Strong |
| SDK wrapper | Application | Direct clients and raw credentials are blocked or the mode is explicitly advisory. | Advisory to moderate |
| Sandboxed runtime | Runtime | Filesystem, network, process, credentials, and tools are unavailable outside governed paths. | Runtime-enforced when complete |
What prevents the agent from producing this effect without crossing Anthesis?
Current maturity
Use precise claims for each layer.
Public materials distinguish deterministic evaluation, bounded reference execution, and broader production assurance.
Runnable now
Deterministic public validation
Signed evaluator acquisition, 7 canonical scenarios, 9 packs / 27 scenarios, 24 inference-integrity scenarios, reports, controlled mismatch exercises, and reproducible evidence.
Reference integration
Bounded governed-agent execution
Dubnium demonstrates one composition with exact authorization binding, approval-gated constrained tools, and runtime evidence.
In development
Broader production assurance
Additional enforcement profiles and stronger live inference-integrity capture, replay, independent verification, containment, and recovery.
Trust boundary: Governance Lab demonstrates deterministic contract behavior over synthetic declarations and recorded evidence. It does not prove universal effect enforcement or live non-bypassability. Those guarantees depend on the actual integration environment.
Learn more
Choose the depth that matches your question.
Run Governance Lab
Use the short stakeholder walkthrough or reproduce every current public proof surface.
Inspect bounded execution
Follow the reference decision, approval, execution, and evidence path.
Read the project brief
See the problem, proof model, integration modes, maturity, and adoption path in one concise document.
Read the engineering notes
Long-form context on the problem, architecture, and public governance proof. This project site remains authoritative for current product status.
24 Scenarios That Prove Your AI Agent Follows Rules →
Anthesis Update: Memory, Governance, and Beyond the SDLC →
AI SDLC: automating the grind →
All Anthesis posts →