Generated Apr 25, 2026, with manifest, checksum ledger, verifier wheel hash, and local verification path.
G‑14 proof-carrying verification layer
Check the evidence. Gate the action. Prove the decision.
G‑14 builds external verification layers for consequential AI: systems that decide whether proposed actions, reconstructions, and conclusions are admissible before they touch robots, tools, records, workflows, or scientific decisions.
PreserveCorrectHoldBlockReleaseProve
Action Proposal Intake Boundary armed
Models
Agents
Robots
Lab Systems
Enterprise AI
Proposed Action
External Control LayerAdmission decisions before effectPPreserve
CCorrect
HHold
BBlock
RRelease
VProve
Governed Execution
Proof Receipt: signed, retained, replayable
Published proof packetEvidence buyers can verify locally.
Review proof evidence Verify the pack yourself Immediate allow, hold-release, hold-block, timeout-safe-state, and replay-seed allow packets are expected to verify.
Unsigned, tampered, missing cross-view, missing inclusion, and signer-topology failures reject as expected.
Packets, receipts, stale and replay rejections, public keyset, checksums, and offline verifier material are pinned.
Built for technical review; certification and production deployment claims remain customer and environment specific.
One control pattern, multiple domains
G‑14 builds proof-carrying verification layers for consequential AI.
Check evidence before a robot command, enterprise action, or agent conclusion crosses its boundary. Every decision can require repair, stop unsafe paths, and leave an auditable receipt that survives outside the model.
Physical AI: TargetLock
Robot and automation systems propose physical actions. G‑14 checks authority, evidence, zone state, and release conditions before the effect channel acts.
Proof receipt for preserve, correct, hold, block, release, and outcome.Agent communication and scientific workAgent collectives reconstruct facts, pass intermediate claims, and converge on conclusions. G‑14 research treats communication as a certifiable process.
Semantic receipt for proceed, re-query, repair, or reject-with-witness.Enterprise AI actionAgents call tools, write records, trigger workflows, and prepare decisions. G‑14 gates the action path without requiring a new foundation model.
Action receipt for proposal, policy, evidence, decision, release path, and verifier result.Where admission matters
Customers do not onboard by naming models. They start with the places AI can touch the business and the actions that cannot be left to probabilistic output alone.
Life Sciences and Pharma
Govern SOP drafts, batch-related documents, deviations, and regulated workflow releases.
Map action riskRobotics and Automation
Admit robot movements, grip actions, AMR routes, and human-shared zone entry.
Inspect boundaryIndustrial Systems
Hold or block machine-adjacent commands until operating evidence is present.
Review pathEnterprise AI
Control tool calls, API writes, record changes, deployments, and agent workflows.
Trace authorityDefense and Mission Systems
Create proof-bearing gates for high-consequence connected and disconnected action paths.
Build proofFinance and Operations
Review refunds, payments, vendor changes, case decisions, and other material actions.
Define permitsSoftware Delivery
Let coding agents open work while merges, secrets, and production changes require admission.
Connect SDKThe infrastructure of trust for AI action
G‑14 is a control surface and a runtime boundary. It accepts action proposals and agent-derived claims, evaluates authority and evidence, controls consequence, and retains proof.
Policy Engine
Permit-to-act rules define actor, verb, resource, authority, evidence, boundary, and recovery.
Trust and Signals
Context, identity, system state, sensor evidence, and business signals enter as typed inputs.
Admission Engine
The proposed action is evaluated against deterministic policy before it can cross the boundary.
Execution Control
The effect channel requires the G‑14 decision before action becomes real.
Proof and Evidence
Receipts bind proposal, rule, evidence, decision, signer, timing, outcome, and retention path.
Audit and Replay
Investigators can replay the chain and export an evidence package for review.
Proof receipt vault
Every consequential action gets a replayable record.
The proof object is not an audit log afterthought. It is the durable record of who proposed the action, what G‑14 decided, under which permit, with what evidence, and what happened after release or block.
- Base model or agent proposal preserved before execution.
- Policy version and admission decision retained with timing.
- Evidence checks, missing signals, corrections, holds, blocks, and releases recorded.
- Human authority, escalation, revocation, and recovery path captured when required.
- Receipt designed for audit, incident review, dispute support, and customer evidence export.
Hold, operator review, releaseamr-evidence-packet-9f34c049c4bb
t+200.0sProposedzone_authorize: AMR amr-12 requests entry to zone-g3, warehouse-east.
t+200.0sHeldStructural lane holds the action; dispatch suppressed; robot keeps safe state hold_position.
t+220.0sReviewedOperator allow: "Zone clear; camera and scanner evidence checked."
t+220.0sReleasedDispatch executed through the governed fleet channel (operator_allow).
t+221.0sSealeded25519 product signature over the canonical packet body.
A signed receipt from the downloadable AMR v0 proof pack. Times are seconds on the demo-run clock. Verifier result: pass; 225/225 checks, signature valid.
unzip g14-amr-customer-proof-pack-v0-20260425.zip
pip install --no-index --find-links verifier-wheelhouse agent-control-runtime
python tools/verify_pack.pyDownload the pack (5.2 MB) · sha256 b00b2ce5298e2795… · Read the verify guide
Governance Mapping Center
Export what the boundary produced, not a paperwork substitute.
NIST AI RMFGovern, Map, Measure, and Manage evidence produced by the action boundary.
NIST CSF 2.0Govern, Identify, Protect, Detect, Respond, and Recover mappings for AI action paths.
ISO/IEC 42001AI management-system evidence for risk controls, oversight, and improvement.
SOC 2 and ISO/IEC 27001Control evidence for access, change, auditability, incident, and trust services review.
ISA/IEC 62443Industrial-zone and control-system alignment for high-consequence environments.
EU AI Act readinessTraceability, oversight, documentation, and operational-control evidence where applicable.
Standards, certification, and authorization references describe evidence mappings and readiness paths, not completed certifications unless expressly stated in a signed customer or assessor document.
Private proof console
Bring one workflow. See what G‑14 would admit, hold, correct, block, release, and prove.
The lowest-friction enterprise start is a private proof on a real action path. Replay historical runs, shadow live proposals, evaluate semantic or physical receipts, then move toward assist and control.