DoD Responsible AI Strategy and Implementation Pathway

What the Strategy Commits To

The June 22, 2022 Strategy and Implementation Pathway (S&IP) is an internal DoD policy statement, not a statute, that operationalizes the five Ethical Principles for AI adopted Feb. 24, 2020 — Responsible, Equitable, Traceable, Reliable, Governable. It does so through six foundational tenets: RAI governance clarifying roles among OUSD(R&E), OUSD(A&S), the DoD CIO and the CDAO; warfighter trust via calibrated reliance and T&E/V&V — the kind of assurance machinery that broader governance scholarship warns is still underdeveloped, lacking 'the mechanisms and institutions to prevent misuse and recklessness' ¹; integration of RAI across the acquisition lifecycle; requirements validation through JCIDS gating; a responsible AI ecosystem covering data sourcing and vendor disclosure; and an AI workforce training mandate. The 'Reliable' principle commits capabilities to 'explicit, well-defined uses' subject to testing within those uses, while 'Traceable' commits to documentation and explainability so that relevant personnel understand the technology, its development, and its operational methods — an obligation that maps onto the explainability needs scholarship identifies for accountable public-sector AI ².

Standing Relative to Binding Law

The S&IP is in force as DoD-wide guidance but carries no external statutory enforceability: it binds the Department through its own management chain, paired with a CDAO-maintained RAI Toolkit of templates and sample contract language, rather than conferring rights on affected parties. It is deliberately distinct from DoDD 3000.09 (Autonomy in Weapon Systems), the directive that governs lethal-autonomy decisions and was separately updated in January 2023; the S&IP routes catastrophic and mission risk through the 'Reliable' principle and JCIDS validation while leaving LAWS-specific decisions to that directive (U.S. Department of Defense 2023). This self-regulatory posture exemplifies the national-security pattern that civilian regimes externalize: where the EU AI Act, Regulation (EU) 2024/1689, embeds security exemptions that critics argue make oversight 'extremely difficult' ³ — exceptions that widened through negotiation to produce 'double standards for fundamental rights protection' ⁴ — the DoD instead writes its own internal framework rather than carving out from a binding civilian baseline.

Critiques and Coverage Gaps

The S&IP's principle-and-tenet architecture leaves several governance functions only implicit. Foundation-model-specific duties are not addressed directly; they 'flow through' Toolkit guidance regardless of architecture — a gap sharpened by definitional instability in the very category ⁵ and by the fact that autonomous generation strains legal categories of authorship, accountability, and control that any procurement assurance must rest on ⁶, compounded by training-data leakage risks ⁷. Compute-threshold reporting is likewise implicit, surfacing only through standard acquisition channels, even as scholarship shows such thresholds are vulnerable to compute-reducing enhancement techniques ⁸. Most acutely, redress under 'Governable' reaches only operator-facing disengagement, not affected-civilian remedy — a deficit relative to empirically grounded contestability needs ⁹ and the explainability-contestability link for public-sector AI ².

Adoption Trajectory and Outlook

Since adoption the S&IP has functioned as the connective tissue between the 2020 Ethical Principles and operational practice, with the CDAO consolidating governance authority and maintaining the RAI Toolkit as the living delivery vehicle — meaning the instrument's real bite evolves through toolkit revision rather than amendment of the strategy text. Its catastrophic-risk handling, channeled through the 'Reliable' principle and JCIDS, remains partial against scholarship urging precautionary state obligations to regulate AI's extreme tail risks ¹⁰ and the distinction between decisive and accumulative existential risk ¹¹; nuclear-domain AI dynamics further test its mission-bounded frame ¹². As a voluntary internal pathway it cannot supply the binding audit-and-halt machinery proposed for international AI safety governance ¹³, leaving its trajectory dependent on continued executive commitment.

Cross-jurisdiction comparison

How peer instruments treat the topics DoD Responsible AI Strategy and Implementation Pathway governs.

°= industry self-imposed voluntary framework. Comparing a voluntary code's "governs" tint with a binding regulation's "governs" tint flattens the legal-force distinction; use the instrument-page banner for the operative status of each.

Does this instrument’s approach work? — the social-science evidence

Aggregated over the 6 topics this instrument governs: whether each harm is empirically real, and whether the peer-reviewed evidence shows governance reduces it. The badge is the epistemic status of the evidence— “thin”/“absent” efficacy evidence is itself a finding (the “second silence”). Each epistemic-status label is Policy Window's editorial assessment of the cited evidence base (a structured classification), not a verdict any single source issues.

Of the 6 governed topics with a social-science evidence review, evidence that governance reduces the harm is established for 0, contested for 0, thin for 0, and absent for 6 — for most, no replicated study yet shows this instrument's approach works (the "second silence").

• Catastrophic & Existential Risk

  • Is the harm real?evidence: contested

    The catastrophic-uplift premise is genuinely contested: the empirical uplift studies that exist find current frontier models add little. RAND's red-team study found no statistically significant difference in the viability of bioweapon-attack plans produced with vs. without LLMs (Mouton, Lucas & Guest 2024), and OpenAI's 100-participant trial found GPT-4 gave at most a mild, non-significant accuracy uplift (mean +0.88 out of 10 for PhD experts, +0.25 for students; Patwardhan et al. 2024). Honest caveat: the harm is forward-looking, not yet observed — expert opinion on the catastrophic tail is sharply split (median AI researcher puts ~5% on extremely-bad/extinction outcomes, mean ~9-16% across differently-framed questions, n=2,778; Grace et al. 2024), and forecasters underestimated how fast risk-relevant capabilities (e.g. virology troubleshooting) actually arrived (Forecasting Research Institute 2025), so the relevant capabilities are a moving target rather than a settled magnitude.

    Sources: Mouton, Lucas & Guest 2024 (RAND RR-A2977-2, Operational Risks of AI in Large-Scale Biological Attacks: Results of a Red-Team Study); Patwardhan et al. 2024 (OpenAI, Building an Early Warning System for LLM-aided Biological Threat Creation); Grace et al. 2024 (Thousands of AI Authors on the Future of AI, arXiv:2401.02843); Forecasting Research Institute 2025 (Forecasting LLM-enabled Biorisk and the Efficacy of Safeguards)

  • Does governance work?evidence: absent

    There is essentially no impact evidence that catastrophic-risk governance reduces catastrophic risk, and structurally there cannot yet be: the harm is a low-probability civilisational tail event, so no controlled trial or before/after evaluation of a realised catastrophe is possible. The dominant instruments are recent, voluntary developer frameworks (Anthropic's Responsible Scaling Policy 2023; OpenAI's Preparedness Framework 2023) built on if-then capability thresholds the developers themselves describe as speculative and qualitative rather than validated risk thresholds. The closest evidence is adjacent and indirect: trained-in deceptive behaviours can persist through standard safety training (Hubinger et al. 2024) — a demonstration that current mitigation may be insufficient, not that any governance regime works — and Anthropic's documented loosening of earlier commitments (RSP 2025 dropped the original pledge to define higher-tier ASL evaluations before developing the corresponding models) illustrates that even the strongest voluntary regimes lack external enforcement or measured efficacy.

    Sources: Anthropic 2023 (Responsible Scaling Policy); OpenAI 2023 (Preparedness Framework); Hubinger et al. 2024 (Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training, arXiv:2401.05566); Hendrycks, Mazeika & Woodside 2023 (An Overview of Catastrophic AI Risks, arXiv:2306.12001)

• Compute-Threshold Reporting

  • Is the harm real?evidence: contested

    Whether training-compute (FLOP) is a defensible proxy for governance-relevant capability is genuinely contested in the literature. The strongest empirical pressure against it is algorithmic efficiency: Ho, Besiroglu, Erdil et al. (2024) estimate the compute needed to reach a fixed language-model performance level has halved roughly every eight months (95% CI ~5-14 months, i.e. ~3x/year), so any static FLOP-to-capability mapping decays quickly; Hooker (2024) argues FLOP measures operations rather than end-performance, since techniques such as fine-tuning, retrieval, chain-of-thought and tool use can add large capability gains without proportional training compute, and Ord (2025) shows inference-time scaling further decouples deployed capability from training compute. Honest caveat: defenders (Heim & Koessler 2024; Pilz, Heim & Brown 2025) note compute remains the most quantifiable, externally verifiable, and ex-ante measurable correlate of frontier capability currently available, while themselves conceding it is an imperfect proxy that should not be used in isolation — the disagreement is about durability and precision, not whether any correlation exists.

    Sources: Ho, Besiroglu, Erdil, Owen, Rahman, Guo, Atkinson, Thompson & Sevilla 2024, Algorithmic progress in language models, NeurIPS 2024 (arXiv:2403.05812; Epoch AI); Hooker 2024, On the Limitations of Compute Thresholds as a Governance Strategy (arXiv:2407.05694); Ord 2025, Inference Scaling Reshapes AI Governance (arXiv:2503.05705); Heim & Koessler 2024, Training Compute Thresholds: Features and Functions in AI Regulation (arXiv:2405.10799); Pilz, Heim & Brown 2025, Increased Compute Efficiency and the Diffusion of AI Capabilities (AAAI 2025; arXiv:2311.15377)

  • Does governance work?evidence: absent

    There is no rigorous evidence that compute-threshold reporting reduces harm or achieves its stated aim, because the regimes have not produced an evaluable record. The US 10^26-FLOP reporting obligation (Executive Order 14110, invoking the Defense Production Act) was revoked on 20 January 2025 (by EO 14148) before its recurring binding reporting rule was finalized — the implementing BIS notice of proposed rulemaking (Sept 2024) never took effect, so no durable reporting record materialized; and the EU AI Act's 10^25-FLOP systemic-risk obligations for general-purpose models only became applicable on 2 August 2025 (with transitional periods into 2027), so no outcome evaluation yet exists. Moreover the 10^25 figure is a rebuttable presumption sitting alongside qualitative high-impact criteria (Art. 51(1)(a) and (2), rebuttable under Art. 52(2)), not a validated risk cutoff. The closest analogue is the broader regulatory-disclosure-mandate literature (Fung, Graham & Weil 2007), which documents that transparency policies' effects on outcomes are highly heterogeneous and frequently ineffective or counterproductive absent enforcement and downstream use — implying that the reporting trigger working as intended is an open empirical question, not a documented result.

    Sources: U.S. Executive Order 14110 (2023), Sec. 4.2 (10^26 FLOP, Defense Production Act); revoked by Executive Order 14148 (Jan 20, 2025); EU AI Act, Reg. (EU) 2024/1689, Art. 51 (10^25 FLOP systemic-risk rebuttable presumption; applicable Aug 2, 2025); Fung, Graham & Weil 2007, Full Disclosure: The Perils and Promise of Transparency (Cambridge University Press)

• Foundation Models / GPAI

  • Is the harm real?evidence: contested

    Whether the foundation-model category maps to a coherent capability/risk tier is genuinely contested. The original case rests on scale-driven 'emergent abilities' that appear unpredictably above a size threshold (Wei et al. 2022; Ganguli et al. 2022 documented capabilities that are smoothly predictable in aggregate loss yet locally surprising), but Schaeffer, Miranda & Koyejo (2023, a NeurIPS Outstanding Paper) showed many 'emergent' jumps are artefacts of discontinuous metrics and dissolve under linear/continuous scoring — implying capability scales more smoothly than a sharp tier would suggest. Honest caveat: this is a live empirical disagreement about measurement, not a settled finding either way, and compute (the regulatory proxy) is an imperfect stand-in for capability or risk regardless of which side is right.

    Sources: Wei et al. 2022 (Emergent Abilities of Large Language Models, TMLR; arXiv:2206.07682); Schaeffer, Miranda & Koyejo 2023 (Are Emergent Abilities of Large Language Models a Mirage?, NeurIPS 2023, Outstanding Paper; arXiv:2304.15004); Ganguli et al. 2022 (Predictability and Surprise in Large Generative Models, ACM FAccT; DOI 10.1145/3531146.3533229)

  • Does governance work?evidence: absent

    There is no impact evaluation showing that GPAI/foundation-model governance reduces harm — the rules are too new (EU AI Act GPAI obligations and the 10^25-FLOP systemic-risk presumption only began binding on 2 August 2025) and the central regulatory lever is itself contested: Hooker (2024) argues compute thresholds are a shortsighted proxy because compute does not reliably track capability or risk, and the thresholds already diverge across jurisdictions (EU 10^25 vs. the now-rescinded US EO 14110's 10^26 operations, rescinded 20 January 2025). The mandated mitigation methods also lack validated efficacy: model evaluation and red-teaming face well-documented coverage limits and an 'audit gap' in the survey/position literature (behavioural testing cannot establish the absence of untested failure modes), and adversarial red-teaming repeatedly defeats deployed safeguards — the UK AI Safety Institute reports finding universal jailbreaks for every frontier system it has tested, and a large public agent-injection competition elicited policy violations across all 22 frontier models tested from ~1.8M attacks (Zou et al. 2025). Even compliant evaluation therefore cannot yet certify the safety the rules demand. (Caveat: this is an absence-of-evidence claim — no efficacy study has been done — not evidence the rules are ineffective.)

    Sources: Hooker 2024 (On the Limitations of Compute Thresholds as a Governance Strategy, arXiv:2407.05694); EU AI Act Arts. 51 & 55 (GPAI systemic-risk presumption, 10^25 FLOP; binding 2 Aug 2025); US EO 14110 (10^26-operation reporting threshold, rescinded 20 Jan 2025 by EO 14148); Zou et al. 2025 (Security Challenges in AI Agent Deployment: Insights from a Large Scale Public Competition / Gray Swan Arena, arXiv:2507.20526 — 22 frontier agents, ~1.8M attacks); UK AI Safety/Security Institute, Frontier AI Trends Report (universal jailbreaks for every system tested); METR, Common Elements of Frontier AI Safety Policies (2024)

• National Security Carveouts in AI Regulation

  • Is the harm real?evidence: thin

    That civilian AI-governance instruments carve out national-security uses is black-letter and undisputed (EU AIA Art. 2(3); CoE Framework Convention Art. 3(2) on national-security activities, distinct from Art. 3(4) on national defence; US NSM-25 (Oct. 2024) as the national-security-track instrument fulfilling §4.8 of EO 14110); civil-society legal analysis argues a blanket exclusion is harder to square with a necessity-and-proportionality approach than a qualified one (Korff/ECNL 2022; Vogiatzoglou 2024). But whether the carveout itself produces concrete unredressed harm is empirically under-observed almost by construction — the secrecy it confers suppresses the very evidence needed to measure it. The closest analogue, national-security deference in the courts, shows the mechanism is real (the FISC granted all but eleven of 33,900 applications 1979-2012, a 99.97% approval rate; Sinnar 2022 documents downstream harms to securitized communities), yet Clarke (2014) shows that lopsided ex parte approval rates alone do not prove rubber-stamping, because rational case selection and pre-vetting produce similar rates in ordinary Title III wiretaps (99.93%) and delayed-notice warrants (99.6-99.8%) — so the magnitude of harm attributable to the carveout, as opposed to the legitimate secrecy of the domain, remains genuinely contested.

    Sources: Korff 2022 (ECNL Opinion on the implications of the exclusion of national security from AI legislation, Oct. 2022); Sinnar 2022 (Harvard Law Review Forum 136:59, 'A Label Covering a "Multitude of Sins": The Harm of National Security Deference'); Clarke 2014 (Stanford Law Review Online 66:125, 'Is the Foreign Intelligence Surveillance Court Really a Rubber Stamp?'); EPIC FISC statistics 1979-2012

  • Does governance work?evidence: absent

    There is no impact evaluation showing that any specific design of the national-security carveout — categorical exclusion versus parallel governance track versus civilian-compliance-with-override — measurably improves oversight or reduces harm relative to the alternatives; the question is argued doctrinally (Vogiatzoglou 2024; Korff/ECNL 2022) but has never been tested empirically. The closest analogue evaluation literature is on the parallel-track model already in use for intelligence surveillance (the FISC / FISA oversight regime), and even there the evidence that the mechanism delivers effective scrutiny is itself contested rather than established (Clarke 2014; Sinnar 2022). No direct evaluation exists because the carveouts are recent (EU AIA 2024, CoE Framework Convention 2024, US NSM-25 2024), enforcement actions are by design non-public, and private parties typically lack standing to challenge a specific exempt deployment — the structural features that make the harm hard to observe also make the governance impossible to evaluate.

    Sources: Vogiatzoglou 2024 (Verfassungsblog, 'The AI Act National Security Exception: room for manoeuvres?', 9 Dec. 2024); Korff 2022 (ECNL Opinion, exclusion of national security from AI legislation); Clarke 2014 (Stanford Law Review Online 66:125); Sinnar 2022 (Harvard Law Review Forum 136:59)

• Individual Redress

  • Is the harm real?evidence: established

    The premise behind redress — that affected people lack meaningful recourse against automated decisions — is real, but the flagship instrument is weaker than commonly assumed. Wachter, Mittelstadt & Floridi (2017) show GDPR creates only a limited 'right to be informed,' not a binding 'right to explanation' of specific decisions; and controlled work finds the explanations actually delivered do not measurably improve lay decision accuracy over showing the bare AI prediction (Alufaisan et al. 2021; and a 2022 meta-analysis by Schemmer et al. — screening 393 articles down to 9 in the final analysis — reports 'no effect of explanations on users' performance compared to sole AI predictions,' even though XAI overall had a positive effect). Honest caveat: the legitimacy/dignity value of being heard is empirically well established in the procedural-justice tradition even where outcome accuracy is unchanged, so 'redress fails' depends on which aim is measured.

    Sources: Wachter, Mittelstadt & Floridi 2017 (International Data Privacy Law 7(2):76); Alufaisan, Marusich, Bakdash, Zhou & Kantarcioglu 2021 (Proceedings of the AAAI Conference on AI 35(8):6618); Schemmer, Hemmer, Nitsche, Kühl & Vössing 2022 (AAAI/ACM AIES '22, meta-analysis)

  • Does governance work?evidence: absent

    There is no rigorous impact evaluation showing that mandated redress mechanisms (right-to-explanation, appeal, human-in-the-loop review) actually reduce erroneous or unfair automated decisions — the evidence that the rule works is itself missing. The closest experimental analogues are discouraging: explanations increase humans' acceptance of AI recommendations regardless of correctness (Bansal et al. 2021), and algorithm-in-the-loop oversight can introduce racial disparities and exhibit automation bias rather than reliably catching model errors (Green & Chen 2019). The procedural-justice literature (Tyler 1990; Lind & Tyler 1988) robustly supports a legitimacy and compliance benefit of fair process, but it measures perceived fairness, not reduction of the substantive decision harm redress is meant to cure.

    Sources: Bansal, Wu, Zhou, Fok, Nushi, Kamar, Ribeiro & Weld 2021 (CHI '21); Green & Chen 2019 (Disparate Interactions, ACM FAT* '19); Tyler 1990 (Why People Obey the Law, Yale Univ. Press); Lind & Tyler 1988 (The Social Psychology of Procedural Justice, Plenum Press)

• Transparency Obligations

  • Is the harm real?evidence: contested

    Documentation artifacts (model cards, datasheets) are well-specified as proposals and are genuinely adopted, but the empirical premise that mandated disclosure produces meaningful transparency is contested. Selbst & Barocas (2018) argue inscrutability and non-intuitiveness are distinct problems and that disclosing rules does not resolve the latter, and large-scale audits find documentation is sparsely and unevenly completed: a systematic analysis of 32,111 Hugging Face model cards (Liang et al. 2024) found environmental-impact, limitations and evaluation sections least often filled, and Bhat et al. (2023, 45 practitioners) found a substantial gap between the documentation proposal and actual practice. Honest caveat: the documentation frameworks themselves are real and adopted, so the dispute is about whether disclosure conveys decision-relevant information, not whether the artifacts exist.

    Sources: Selbst & Barocas 2018 (Fordham Law Review 87:1085-1139); Liang et al. 2024 (Nature Machine Intelligence, s42256-024-00857-z, 'Systematic analysis of 32,111 AI model cards'); Bhat et al. 2023 (CHI '23, 'Aspirations and Practice of ML Model Documentation', DOI 10.1145/3544548.3581518); Mitchell et al. 2019 (FAccT, Model Cards for Model Reporting); Gebru et al. 2021 (CACM 64(12):86-92, Datasheets for Datasets)

  • Does governance work?evidence: absent

    There is no rigorous impact evaluation showing that AI transparency mandates (model cards, training-data summaries) measurably reduce bias, misuse or accidents — the central regulatory assumption is empirically untested, partly because flagship mandates like EU AI Act Art. 53(1)(d) GPAI training-data summaries are only subject to AI Office enforcement/verification from 2 August 2026 (the obligation itself began 2 August 2025 for new models). The closest analogue, mandated consumer disclosure, shows small and context-dependent effects: Bollinger, Leslie & Sorensen (2011) found mandatory calorie posting cut average calories per transaction by about 6%, while Loewenstein, Sunstein & Golman (2014) review evidence that disclosure effects are frequently diminished or even reversed by limited attention and often change provider rather than recipient behavior. These are analogues, not AI studies; no study demonstrates that AI transparency disclosure achieves its stated downstream safety aims.

    Sources: Bollinger, Leslie & Sorensen 2011 (AEJ: Economic Policy 3(1):91-128); Loewenstein, Sunstein & Golman 2014 (Annual Review of Economics 6:391-419, 'Disclosure: Psychology Changes Everything'); EU AI Act Art. 53(1)(d) GPAI training-data summary (obligation from 2 Aug 2025; AI Office enforcement from 2 Aug 2026)

Related instruments

• EU AI Act · EU
• Council of Europe Framework Convention on AI · council_of_europe
• Italy Law No. 132/2025 on Artificial Intelligence (Legge 23 settembre 2025, n. 132) · IT