Audience: CISOs, VP Engineering, GRC teams, Board AI committees, C3PAOs, and Notified Bodies evaluating frontier AI governance readiness.
On September 29, 2026, Google, Meta, NVIDIA, OpenAI, Anthropic, and xAI signed the White House Accord on Super Intelligence, described as "morally binding" with no statutory penalties. The Accord outlines four layers of controls and audits that signatories commit to implementing.
Board committees and external auditors are expected to use the four-layer structure as their de facto governance checklist. Organizations that can demonstrate four-layer compliance with cryptographic evidence are positioned ahead of those relying on internal documentation alone.
For context, other jurisdictions have enacted binding AI legislation. Connecticut's SB 2 (signed 2024) imposes enforceable obligations on AI deployers. The Accord's voluntary structure does not diminish the evidentiary value of cryptographic witness anchors; it increases the importance of demonstrating compliance proactively.
The Joint Commitment on Frontier Responsibilities (Sep 29, 2026) outlines four layers of controls:
| Layer | Commitment |
|---|---|
| Layer 1 | "Implement robust internal controls to monitor capabilities... ensure models do not hack or access technical systems in unintended ways." |
| Layer 2 | "Empower an internal team to ensure all controls, monitoring, and detection are operating as intended, and issues are remediated." |
| Layer 3 | "Partner with an independent external auditor or evaluator to carry out independent assessments of whether controls... are operating as intended." |
| Layer 4 | "Designate an independent committee of the board... to receive reports from internal teams and external auditors." |
Source: Joint Commitment on Frontier Responsibilities, signed Sep 29, 2026.
Layer 1 requires robust controls to record whether frontier models access systems beyond their declared scope. The following SWT3 procedures witness Layer 1 control operations:
| Procedure | Title | What It Witnesses |
|---|---|---|
| AI-ACC.1 | Resource Access Scope Witnessing | Records whether agent resource access stayed within declared scope |
| AI-SHELL.1 | Runtime Containment Attestation | Records whether sandbox policies operated during agent execution |
| AI-MCP.2 | Tool Integrity Attestation | Records whether tool schemas remained unmodified during sessions |
| AI-MCP.6 | Elicitation Consent Witnessing | Records whether agents solicited information beyond authorized scope |
| AI-SEC.1 | Security Perimeter Status | Records security boundary enforcement status during operation |
| AI-GRD.1 | Guardrail Evaluation | Records whether output filtering operated during inference |
Full procedure specifications: swt3.ai/registry
Layer 2 requires an internal team to verify that controls and monitoring are operating as intended. The following SWT3 procedures witness Layer 2 monitoring operations:
| Procedure | Title | What It Witnesses |
|---|---|---|
| AI-DENSITY.1 | Witnessing Coverage Density | Records whether monitoring operated at the declared rate |
| AI-DRIFT.1 | Configuration Drift Baseline | Records whether model behavior remained within declared parameters |
| AI-DRIFT.2 | Consequence-Mapped Drift Response | Records that automated remediation operated when drift was detected |
| AI-INCIDENT.1 | Incident Lifecycle Chain | Records that detection, containment, and remediation steps occurred |
| AI-INF.1 | Inference Logging | Records prompt and response hashes for continuous audit trail |
Layer 3 requires partnership with an independent external auditor to assess whether controls operated as intended. The following SWT3 procedures witness Layer 3 assessment operations:
| Procedure | Title | What It Witnesses |
|---|---|---|
| AI-AUDIT.1 | Audit Log Integrity | Records that audit trails were not modified after generation |
| AI-ASSESS.1 | Champion-Challenger Assessment | Records that model comparison evaluations were conducted |
| AI-EXPL.1 | Explainability Metadata | Records that model decisions can be examined by external assessors |
Layer 4 requires an independent board committee to receive reports from internal teams and external auditors. The following SWT3 procedures witness Layer 4 governance operations:
| Procedure | Title | What It Witnesses |
|---|---|---|
| AI-GOV.1 | Governance Policy Attestation | Records that AI governance policies were reviewed and approved |
| AI-EMRG.1 | Emergency Override Lifecycle | Records that critical interventions were authorized for board review |
Runtime containment tools generate operational evidence: OCSF audit logs, sandbox traces, and policy enforcement records. Examples include OpenShell, gVisor, Kata Containers, and Firecracker, all of which are supported runtime types in the AI-SHELL.1 procedure.
SWT3 records that evidence as cryptographic witness anchors with verifiable fingerprints. Runtime containment generates the evidence. SWT3 records it as independently verifiable witness anchors.
These layers are complementary. The runtime tool proves a sandbox was active. The SWT3 anchor proves that enforcement was witnessed, timestamped, and recorded in an immutable chain that external auditors can verify without accessing the runtime environment.
Configure the SI Accord 2026 governance profile in one command:
swt3 init --profile si-accord-2026This generates a .swt3.yaml configuration file with all 16 Accord-mapped procedures pre-configured.
Full procedure specifications: swt3.ai/registry. Integration documentation: sovereign.tenova.io/docs.
Start with the Open tier (free). Board Audit Package export is available at Pro ($799/mo).
| Accord Layer | Procedure | Title |
|---|---|---|
| Layer 1 | AI-ACC.1 | Resource Access Scope Witnessing |
| Layer 1 | AI-SHELL.1 | Runtime Containment Attestation |
| Layer 1 | AI-MCP.2 | Tool Integrity Attestation |
| Layer 1 | AI-MCP.6 | Elicitation Consent Witnessing |
| Layer 1 | AI-SEC.1 | Security Perimeter Status |
| Layer 1 | AI-GRD.1 | Guardrail Evaluation |
| Layer 2 | AI-DENSITY.1 | Witnessing Coverage Density |
| Layer 2 | AI-DRIFT.1 | Configuration Drift Baseline |
| Layer 2 | AI-DRIFT.2 | Consequence-Mapped Drift Response |
| Layer 2 | AI-INCIDENT.1 | Incident Lifecycle Chain |
| Layer 2 | AI-INF.1 | Inference Logging |
| Layer 3 | AI-AUDIT.1 | Audit Log Integrity |
| Layer 3 | AI-ASSESS.1 | Champion-Challenger Assessment |
| Layer 3 | AI-EXPL.1 | Explainability Metadata |
| Layer 4 | AI-GOV.1 | Governance Policy Attestation |
| Layer 4 | AI-EMRG.1 | Emergency Override Lifecycle |