D16: AI Governance and Resilience Aspects in CODECO v2.0

Tool: https://gitlab.eclipse.org/eclipse-research-labs/codeco-project/carg

Executive Summary:

Deliverable D16 of CODECO (CODECO AI Governance and Resilience Aspects, M38,
February 2026) corresponds to an update of D15 of WP5 (Experimentation Framework,
Demonstrations and AI Data Governance and Resilience Testing), T5.1 (AI Data Governance
and Resilience Aspects).
Task T5.1 addressed the governance and resilience implications of embedding AI-driven
mechanisms into orchestration processes across the CEI continuum. In distributed and
federated environments such as those targeted by CODECO, AI is not an isolated application
but an enabling layer influencing workload placement, resource allocation, and system
adaptation. This integration introduces governance, resilience, transparency, and
sustainability challenges that must be addressed systematically.
D16 consolidates and operationalizes the CODECO AI Governance and Resilience framework
(CARG) introduced in D15. It documents its application to all six CODECO use-cases as well
as to the core AI-integrating components of the CODECO OSS framework, in particular the
Privacy-preserving Decentralized Learning (PDLC) module and its interactions with other
CODECO components.
The Web-based CARG tool described in this report provides a Minimum Viable Product (MVP)
for a reusable assessment instrument. Through CARG, CODECO:
•Assesses robustness and resilience requirements across use-cases and core
components,
•Verifies the framework’s reliability under heterogeneous data conditions and operational
variability,
•Ensures transparency, explainability, and replicability of AI-driven orchestration decisions
and governance processes.
The assessment results demonstrate that transparency and explainability are consistently
embedded across use-cases, while governance maturity is structured but not uniformly
harmonized. Resilience maturity varies across domains, highlighting the importance of
institutionalized stress testing and adversarial validation. Ethical accountability and
sustainability considerations are present but require further formalization in future
deployments, particularly when transitioning toward higher TRL levels or regulated
environments.
D16 also includes a system-level characterization of CODECO components based on
structured surveys with component leaders. This characterization reflects the intended TRL4–
5 maturity of the software and does not constitute a compliance audit. It provides a transparent
account of governance strengths and limitations, positioning the framework realistically for
future exploitation and scaling.
Sustainability is operationalized within CODECO through energy-aware orchestration and the
capability to incorporate CO₂ or carbon-intensity metrics into placement decisions. While
currently implemented as an optimization feature, this capability provides a foundation for
future integration of structured environmental reporting and KPI tracking.
Finally, the CARG framework and its associated tooling remain available in the Eclipse
CODECO GitLab research repository and will be maintained by fortiss. The asset is expected
to evolve, including potential integration of advanced analytics and Generative AI–assisted
interfaces to enhance usability and governance support. In this way, CARG is positioned as a
reusable governance infrastructure component capable of supporting responsible AI-enabled
orchestration beyond the lifetime of the CODECO project.