DoD AI Ethical Principles — ArgusAiTR Compliance

The DoD's five AI Ethical Principles (Adopted 2020) are Responsible, Equitable, Traceable, Reliable, Governable. ArgusAiTR is designed against all five from the ground up — not bolted on. Specific implementation pointers are included so reviewers can audit the mapping by reading the code.

Responsible

"DoD personnel will exercise appropriate levels of judgment and care while remaining responsible for the development, deployment, and use of AI capabilities."

ArgusAiTR posture: The operator owns every engagement decision. The system advises — it does not engage.

Implementation	Where
HITL hard gate: /op/engage_request is the only path that produces an "engage" event	src/argus_aitr/hmi/server.py (POST /op/{action})
The frontend confirm() modal blocks accidental clicks	src/argus_aitr/hmi/static/app.js
No code path sends a CROWS slew command without an operator-recorded OperatorAction	src/argus_aitr/integrate/crows_adapter.py (mock)
Operator handbook defines the responsibility chain	docs/operator-handbook.md

Equitable

"The Department will take deliberate steps to minimize unintended bias in AI capabilities."

ArgusAiTR posture: Threat assessment uses physical kinematics (range, closing speed, size envelope) — not appearance, identity, or demographic attributes. The system ingests no personal data.

Implementation	Where
Threat rules are transparent and auditable; every verdict carries a reasoning string	src/argus_aitr/classify/threat_heuristic.py
Detector classes are flight-object proxies (airplane, bird, kite); no identity features	src/argus_aitr/detect/yolo_detector.py
Range priors are airframe categories, not personal characteristics	src/argus_aitr/range/monocular.py (DEFAULT_CLASS_PRIORS)
Post-award fine-tuning will use UAS imagery from licensed datasets; no biometric data	docs/roadmap.md

Traceable

"AI capabilities will be developed and deployed such that relevant personnel possess an appropriate understanding of the technology, development processes, and operational methods."

ArgusAiTR posture: Append-only structured audit log records every detection, classification, operator action, and CROWS command. Source code is public. Model lineage and training data sources are documented.

Implementation	Where
AuditLogger writes JSONL events for every notable pipeline transition	src/argus_aitr/audit/log.py
Every operator endpoint records an OperatorAction into the audit stream	src/argus_aitr/hmi/server.py
Threat verdicts include a human-readable threat_reasoning field	src/argus_aitr/types.py (TrackedTarget)
Architecture, integration contract, and licensing documented for reviewers	docs/architecture.md, docs/integration-contract.md, docs/licensing.md
Repo is public Apache-2.0; OSS deps are catalogued with licenses	LICENSE, docs/licensing.md

Reliable

"AI capabilities will have explicit, well-defined uses, and the safety, security, and effectiveness of such capabilities will be subject to testing and assurance within those defined uses, across their entire life cycle."

ArgusAiTR posture: Defined use ("aided UAS detection and threat assessment against Group 1/2 UAS at 100–600 m, with mandatory human authorization for engagement"). Failsafe degradation. Test suite covers detection / track / range / perturbation / HMI handlers / failure paths.

Implementation	Where
SystemHealth.aitr_status exposes nominal / degraded / offline with quantitative notes	src/argus_aitr/types.py, src/argus_aitr/pipeline.py
HMI banner flips to "AiTR DEGRADED — REVERT TO MANUAL" automatically on health change	src/argus_aitr/hmi/static/app.js
Pipeline auto-degrades on FPS below floor or repeated detector errors	src/argus_aitr/pipeline.py (_derive_status)
Perturbation simulator exercises the adverse-conditions branches deliberately	src/argus_aitr/stabilize/perturb.py
Tests: 100+ in tests/ covering every module	tests/

Governable

"The Department will design and engineer AI capabilities to fulfill their intended functions while possessing the ability to detect and avoid unintended consequences, and the ability to disengage or deactivate deployed systems that demonstrate unintended behavior."

ArgusAiTR posture: Single-click /aitr/disable instantly takes the AiTR pipeline offline; the operator continues with the raw video. The system never sends CROWS commands autonomously.

Implementation	Where
/aitr/disable and /aitr/enable route flip the pipeline mode in a single call	src/argus_aitr/hmi/server.py
HMI toggle button is always visible and always authoritative	src/argus_aitr/hmi/static/index.html (#aitr-toggle)
Mock CROWS adapter has no autonomous slew path — every command call is operator-attributed	src/argus_aitr/integrate/crows_adapter.py
Audit log records both disable and enable transitions	src/argus_aitr/audit/log.py
Production GVA adapter is stubbed pending governance review of the wire format	src/argus_aitr/integrate/crows_adapter.py (GVAAdapter)

Summary

ArgusAiTR's compliance posture is not an assertion — it is implemented in code that reviewers can read. The audit log is the externally inspectable artifact that ties every principle back to specific events. Post-award we expand coverage with a formal ATO-style verification & validation plan documented in docs/roadmap.md.