Source-linked AI summary
Position: AI Governance Needs ISO-like Interoperability Protocols, Not Just Laws
Azmine Toushik Wasi, Mst Rafia Islam, Mahfuz Ahmed Anik, Taki Hasan Rafi, Md Manjurul Ahsan, Dong-Kyu Chae
TL;DR
AI governance is fragmented across jurisdictions, creating a need for shared technical language beyond laws alone. The paper proposes modular, machine-readable AI risk manifests, concluding that they can standardize cross-border risk communication while preserving national regulatory autonomy.
Problem
AI governance lacks a shared technical language for expressing and operationalizing risk across fragmented national regulatory approaches.
Method
The paper proposes modular, versioned, machine-readable manifests using standardized metrics for bias, energy consumption, data provenance, and regulatory alignment.
Results
The proposed manifests standardize risk-information representation across jurisdictions without harmonizing legal thresholds, enabling a common technical substrate for enforcement.
Takeaways & Limitations
Interoperable manifests could make AI compliance more auditable and automatable while reducing redundant documentation and easing cross-border deployment.
Takeaways & Limitations
Manifests provide a minimum verifiable baseline, but cryptographic attestations cannot ensure that reported evaluations are truthful or representative.
Abstract
from arXiv · showhide
As Artificial Intelligence (AI) systems become deeply integrated into critical global infrastructure, the urgency for robust governance frameworks has intensified. However, current approaches, led by jurisdiction-specific laws, policies, and voluntary frameworks such as the EU AI Act, China's algorithm governance, and the NIST AI Risk Management Framework in the U.S., create a fragmented regulatory landscape. In this position paper, we argue that \textbf{\textit{AI governance must be built not on laws alone, but on ISO-like interoperability protocols that enable standardized, machine-readable risk communication across borders}}. Drawing on the success of the GDPR, which was operationalized through standards like ISO 27001 and Privacy by Design, we propose the development of standardized AI \textit{nutrition labels} containing unified metrics for bias, energy usage, and data provenance to facilitate cross-jurisdictional compliance. These manifests would lower barriers for small and medium enterprises (SMEs), reduce redundant regulatory efforts, and build public trust. The paper addresses concerns that standards may stifle innovation by advocating for modular, versioned protocols designed to evolve in tandem with technological change. Overall, we call for a shift from siloed legal compliance toward interoperable technical conformance, enabling a shared global language for responsible AI deployment.
1. Introduction
Rapid AI adoption in high-stakes domains has intensified systemic risks while regulatory responses remain fragmented across jurisdictions. The paper argues that laws should be complemented by ISO-like technical interoperability protocols that standardize how AI compliance and risk are communicated in practice.
- Motivation: Over 65% of enterprises report adopting AI, increasing the need for oversight as systems scale across healthcare, critical infrastructure, finance, and organizational workflows.The passage links adoption to efficiency gains and new revenue while emphasizing transformative potential alongside systemic risks.
- Problem: AI regulation remains fragmented, with jurisdictions applying sharply divergent philosophies to systems built on largely similar architectures, data, and risk pathways.The contrast ranges from the EU’s risk-based ex ante controls to more market-driven and voluntary approaches in the US.
- Problem: The resulting standardization vacuum is the absence of a shared technical language and interoperable protocols for communicating AI risk across borders.Without harmonized technical standards, the same AI system may receive different risk classifications across jurisdictions.
- Limitations: Technical standards can become non-tariff barriers when proprietary tooling or compliance costs favor incumbents and exclude SMEs or firms from the Global South.The passage identifies market concentration and limited regulatory capacity as risks of poorly designed standardization.
- Contribution: The paper proposes ISO-like technical interoperability protocols alongside laws to support global accountability, trust, and responsible innovation.Laws remain necessary for normative thresholds and enforcement authority, but technical protocols are needed for consistent operationalization across heterogeneous AI systems.
- Contribution: The proposed model assigns laws the definition of acceptable AI behavior and technical standards the specification of how compliance is demonstrated in practice.It preserves domain-specific variation while maintaining stable interoperability primitives for scalable global AI governance.
2. Background: GDPR’s Success Through Standardized Implementation
GDPR’s global influence was strengthened by ISO 27001 and Privacy by Design, which translated legal obligations into standardized implementation practices. However, AI governance requires product-output manifests because AI systems are adaptive, probabilistic, and context-dependent in ways organizational security standards cannot fully capture.
- GDPR’s Success Through Standardized Implementation: GDPR established harmonized privacy obligations with extraterritorial reach and a unified framework across EU member states.Its objective was to strengthen individual control over personal data while creating consistent regulatory obligations.
- GDPR’s Success Through Standardized Implementation: ISO 27001 operationalized GDPR requirements through risk assessment, security controls, documentation, and accountability within an Information Security Management System.The standard provided a practical and internationally recognized framework for implementing lawful, fair, and secure data processing.
- Limits of the GDPR Analogy: The GDPR–ISO 27001 model cannot transfer wholesale to AI because AI systems are opaque, adaptive, probabilistic, and variable across deployment contexts.AI governance must address model behavior, downstream adaptation, emergent risks, and domain-specific performance variability.
- Governance Lesson: ISO 27001 governs organizational information-security processes, whereas AI manifests must govern deployed product outputs without requiring inspection of internal source code.The proposed AI Risk Manifest is compared with SBOMs as a standardized, versioned artifact that travels with a product and enables downstream auditing.
- GDPR’s Success Through Standardized Implementation: GDPR’s effectiveness came from coupling legal authority with interoperable technical standards and design principles, rather than relying on law alone.The section presents standards as instruments for translating abstract obligations into concrete processes, audits, and safeguards.
3. Machine-Readable AI Risk Manifests
The section proposes machine-readable, ISO-like AI Risk Manifests as a shared technical layer for communicating bias, energy use, and data provenance across heterogeneous regulatory regimes. Standardized, versioned, and regulator-aligned schemas could make compliance more comparable and automatable while preserving jurisdiction-specific legal thresholds.
- Concept and motivation: AI nutrition labels are formalized as technical manifests rather than merely human-readable summaries, addressing missing standardized schemas, versioning, and regulatory mappings.Existing initiatives disclose model and data information or summarize deployed tools, but generally lack machine-readable structure and formal connections to governance frameworks.
- Standardized metrics: The proposed minimum metric set covers bias, energy consumption, and data provenance, creating a shared vocabulary without imposing a single definition of risk.The metrics are intended to be reported, compared, and verified across jurisdictions and deployment contexts while leaving legal thresholds heterogeneous.
- Interoperability gap: Machine-readable manifests address the technical fragmentation that prevents risk information from being consistently expressed and interpreted across regional governance regimes.Without a shared schema, cross-border accountability, automated compliance checks, and meaningful comparison remain limited.
- Regulatory alignment: An explicit regulatory-alignment section maps a manifest’s risk profile to frameworks such as EU AI Act risk tiers and NIST AI RMF functions without harmonizing legal thresholds.The crosswalk allows one manifest to serve as a reusable compliance artifact, reducing redundant documentation and easing cross-border deployment.
- Operationalization: Standardized manifests transform compliance from manual interpretation into auditable, automatable workflows that support validation, monitoring, and post-market auditing.Regulatory or organizational systems can ingest the manifests, increasing efficiency while reducing compliance costs.
- Threat model and verifiability: Cryptographic attestations protect the integrity of reported values but do not by themselves eliminate self-reporting bias, selective disclosure, or cherry-picked evaluations.The section distinguishes integrity—whether signed values were altered—from truthfulness, which remains vulnerable to compliance theater.
4. Policy Framework: Linking Regulatory Compliance to Open Standards
AI governance is fragmented across divergent national and regional frameworks, producing overlapping compliance requirements that raise costs and slow cross-border deployment. Open technical standards are proposed as a coordination infrastructure that links regulatory intent to interoperable, verifiable practice without replacing law.
- Fragmented Regulatory Landscape: Divergent EU, Chinese, U.S., and ASEAN approaches create overlapping and inconsistent compliance requirements, increasing costs and slowing cross-border AI deployment.The EU uses binding obligations, China mandates registration and security reviews, the U.S. relies mainly on voluntary guidance, and ASEAN promotes non-binding principles.
- Regulation Reinforced by Open Standards: Open standards can align regulatory oversight with innovation by enabling interoperability across heterogeneous infrastructure.Open Banking uses interoperable APIs under PSD2, while Smart Grid standards support secure coordination across heterogeneous energy systems.
- Mechanisms for Integrating Standards: Regulators may connect compliance to ISO-like standards for high-risk systems and use public-sector procurement to encourage safer system behavior by design.The proposed mechanisms include conformance with ISO/IEC 42001 for AI management systems and procurement requirements, alongside capabilities to decline unsafe actions or exit hazardous states.
- Coordination Infrastructure: These mechanisms support incremental convergence around shared technical representations of AI risk while treating open standards as coordination infrastructure rather than substitutes for law.The framework acknowledges political, economic, and institutional constraints and does not eliminate global regulatory fragmentation.
- Operationalizing Governance: Interoperable technical standards are presented as necessary to translate statutory intent and policy goals into operational, verifiable practice for globally deployed, rapidly evolving AI systems.The proposed AI risk manifest schema is offered as a response to the insufficiency of legal and ethical frameworks alone.
5. Discussion
Jurisdiction-specific frameworks create a standardization vacuum that undermines interoperability, raises compliance costs, and complicates cross-border accountability. The paper therefore proposes machine-readable, modular, and versioned AI risk manifests to translate regulatory intent into implementable, auditable practice and standardize risk communication.
- Problem: Jurisdiction-specific frameworks create a standardization vacuum that undermines interoperability, raises compliance costs, and complicates cross-border accountability.The paper argues that legal authority alone is insufficient for scalable AI governance.
- Problem: Effective AI oversight requires interoperable technical standards that translate regulatory intent into implementable and auditable practice.
- Contribution: The paper proposes machine-readable, modular, and versioned AI risk manifests functioning as standardized nutrition labels for AI systems.The reusable artifacts integrate fairness, energy use, data provenance, and regulatory alignment to support consistent risk communication across jurisdictions.
6. Alternative Views
The section examines objections that interoperability standards may lag behind AI innovation and may advantage incumbent firms while raising barriers for smaller entrants.
- Alternative Views: Critics argue that technical standards will slow AI innovation because they cannot keep pace with rapidly evolving models, data, and deployment practices.This view favors minimizing formal constraints and allowing practices to evolve organically.
- Alternative Views: A related critique holds that global standards disproportionately benefit large technology firms with established compliance capacity while burdening smaller firms and new entrants.The section presents this concern as a risk that standards could entrench incumbents and raise entry barriers.
7. Recommendations and Call to Action
The section calls for concrete actions to move AI governance from fragmented compliance toward interoperable standards. It emphasizes a shared machine-readable risk baseline and incentive-based adoption to support flexible uptake and lower barriers for smaller actors.
- R1: Establish a shared technical baseline for AI risk communication: Policymakers and standards bodies should establish a minimal, reusable, machine-readable baseline for AI risk manifests across regulatory regimes.This recommendation is framed as R1 and is intended to move governance toward interoperability.
- Incentive-based adoption: Governments, funders, and large procurers should reward interoperable AI risk standards through procurement, certification, and research-evaluation criteria.The passage presents these incentives as mechanisms for accelerating standard uptake while preserving flexibility.
- Incentive-based adoption: A single reusable artifact can lower smaller actors’ entry barriers by replacing multiple bespoke requirements.The passage connects this benefit to incentive-based adoption of interoperable AI risk standards.
8. Conclusion
As AI systems become globally deployed and embedded in high-stakes domains, law alone cannot scale as a governance mechanism. The paper therefore calls for ISO-like interoperability protocols and machine-readable AI risk manifests to support standardized, verifiable, cross-border risk communication.
- Conclusion: Law-alone governance mechanisms cannot scale as AI systems become globally deployed, continuously updated, and embedded in high-stakes domains.
- Conclusion: Effective AI governance requires ISO-like interoperability protocols that complement law with standardized, verifiable, cross-border risk communication.
- Conclusion: The paper proposes machine-readable AI risk manifests, described as nutrition labels, to operationalize interoperable AI governance.
A. Detailed Worked Example of AI Nutrition Label
The worked example presents an AI Risk Manifest as a signed, structured, machine-readable object for cross-border exchange and validation. It standardizes risk evidence while leaving jurisdictions authority over thresholds and enforcement actions.
- Manifest structure: The AI Risk Manifest records system identity, intended use, risk tier, applicable regimes, and standardized fairness, performance, energy, and privacy/security metrics.It is designed as a versioned payload that can be exchanged across borders and automatically validated.
- Manifest structure: The example identifies RawModel-1 as an LLM+ranker, version 2.3.1, deployed through an API in EU, US, and BD regions.Its model fingerprint is sha256:7c3b...e91a.
- Risk and use constraints: The manifest classifies recruiter resume triage as high_risk employment decision support with human review required and prohibits fully-automated hiring decisions.It also prohibits use outside declared job families without re-validation.
- Evaluation results: For the heldout EU dataset HR-Triage-2025Q4 with n: 50000, topk_precision@10 is 0.61, auc is 0.79, ood_drop_auc is 0.06, and prompt_injection_pass_rate is 0.93.These fields illustrate how performance and robustness evidence can be represented in the manifest.
- Governance boundary: The manifest standardizes how risk evidence is represented and exchanged while jurisdictions retain authority over applicable thresholds and enforcement actions.This separates interoperable evidence formats from jurisdiction-specific regulatory decisions.
B. Why Technical Schemas Can Succeed Where Political Treaties Fail
Technical interoperability standards can achieve broader practical reach than political treaties because they diffuse through market incentives without requiring prior political alignment. In AI governance, reusable machine-readable risk manifests can reduce cross-border compliance friction and serve as coordination infrastructure.
- Political treaties often stall because they are slow to negotiate, difficult to enforce, and constrained by sovereignty concerns, geopolitical competition, and divergent economic priorities.
- Technical interoperability standards diffuse through market adoption incentives, supply-chain pressure, and procurement requirements, achieving de facto global reach without formal international agreements.
- Cross-border firms have incentives to adopt a single reusable compliance artifact instead of maintaining jurisdiction-specific documentation pipelines.Machine-readable risk manifests can reduce regulatory friction, accelerate market access, and lower legal uncertainty for companies deploying AI across jurisdictions.
- Interoperable AI risk manifests function as coordination infrastructure by allowing firms to comply once and reuse documentation across jurisdictions.This makes partial alignment economically preferable to fragmented compliance approaches.
- Technical schemas can spread through procurement, certification, and contractual norms, as shown by earlier successes in financial reporting, cybersecurity, and supply-chain transparency.
C. Verification, Attestation, and the Limits of Self-Reporting · D. Addressing the Innovation vs. Standards Dilemma · D.1. Counterargument: Standards Will Lag Behind Innovation
The proposed manifest system addresses self-reporting risks through cryptographic attestations and external verification, while standards remain a concern because rigid rules could slow AI progress and disadvantage new entrants.
- C. Verification, Attestation, and the Limits of Self-Reporting: Self-reporting can bias AI documentation when developers disclose favorable metrics, omit unfavorable information, or use unrepresentative test sets.
- C. Verification, Attestation, and the Limits of Self-Reporting: External validation is required to establish truthfulness, because an intact manifest can otherwise become formally complete but substantively unreliable.Third-party verification operationalizes the layer connecting cryptographic integrity with trustworthy claims.
- C. Verification, Attestation, and the Limits of Self-Reporting: Cryptographic attestations bind manifest fields such as model versions, dataset hashes, evaluation metrics, and audit artifacts to signed hashes.Signatures provide immutability and accountability but do not establish that the underlying evaluation is correct.
- C. Verification, Attestation, and the Limits of Self-Reporting: Independent auditors can attest to the same hashed artifacts, allowing regulators and deployers to cross-check claims without rerunning full evaluations.This shifts trust toward the integrity of the verification chain and raises the cost of deception, though it does not eliminate garbage-in risks.
- C. Verification, Attestation, and the Limits of Self-Reporting: Verification is incremental rather than absolute, establishing a minimum verifiable baseline that can support reproducible environments, third-party audits, and regulatory spot checks.The design reduces the feasibility and impact of misconduct without preventing all misconduct.
- D.1. Counterargument: Standards Will Lag Behind Innovation: Critics argue that regulations and standards may stifle innovation by slowing AI advancement, creating entry barriers, and potentially strengthening incumbents.They also contend that rigid rules could hinder AI’s adaptability and learning from new data.
D.2. Response: Modular, Versioned Standards for Agile Evolution · E. Overcoming the “Cold Start” Problem · E.1. A Six-Stage Cold Start Pathway
The paper proposes agile, interoperable AI standards built from modular, versioned, open, and outcome-oriented components rather than prescriptive rules. It outlines a cold-start pathway in which procurement and market incentives drive adoption before regional convergence and eventual formalization.
- D.2. Response: Modular, Versioned Standards for Agile Evolution: Modular standards can evolve independently across bias assessment, data provenance, and energy usage without requiring systemic overhauls.The paper presents modularization as a way to preserve adaptability while avoiding comprehensive revisions when one component changes.
- D.2. Response: Modular, Versioned Standards for Agile Evolution: Versioned, iterative standards use regular updates and feedback loops to remain responsive to emerging AI challenges and opportunities.The proposed process draws inspiration from software development and the Agile Manifesto.
- D.2. Response: Modular, Versioned Standards for Agile Evolution: Open, collaborative standard-setting can improve quality, usability, and security by exposing blind spots, incorporating diverse stakeholder needs, and accelerating adoption.The paper specifically advocates transparent, community-driven processes and open-source participation.
- D.2. Response: Modular, Versioned Standards for Agile Evolution: Outcome-oriented standards should define targets such as fairness or energy efficiency while leaving developers freedom over architectures and mitigation techniques.This approach is intended to avoid unnecessarily prescriptive technical implementation requirements.
- D.2. Response: Modular, Versioned Standards for Agile Evolution: Globally harmonized risk standards can function as enabling infrastructure by providing a common technology stack that reduces redundant work, simplifies cross-border deployment, and levels the playing field.The paper compares this role to standardized network protocols and financial APIs.
- E. Overcoming the “Cold Start” Problem: Partial convergence around a minimal interoperable schema is sufficient for procurement, auditing, and compliance workflows without requiring full global consensus.The proposed framework therefore does not depend on complete international agreement before functioning.
- E.1. A Six-Stage Cold Start Pathway: The six-stage pathway begins with public-sector procurement and proceeds through cloud integration, open-source tooling, regional adoption, ISO/IEC formalization, and modular extension.Government purchasing creates demand, shared tooling lowers compliance barriers, and later extensions accommodate domains such as healthcare, finance, and agentic systems.
- E.1. A Six-Stage Cold Start Pathway: The pathway uses market access and transaction cost reduction to drive convergence without requiring geopolitical consensus as a precondition.The proposed sequence relies on procurement incentives, supply-chain integration, and shared technical infrastructure rather than prior treaty negotiation.