Infusion of Blockchain to Establish Trustworthiness in AI Supported Software Evolution: A Systematic Literature Review

Context: Blockchain and AI are increasingly explored to enhance trustworthiness in software engineering (SE), particularly in supporting software evolution tasks. Method: We conducted a systematic literature review (SLR) using a predefined protocol with clear eligibility criteria to ensure transparency, reproducibility, and minimized bias, synthesizing research on blockchain-enabled trust in AI-driven SE tools and processes. Results: Most studies focus on integrating AI in SE, with only 31% explicitly addressing trustworthiness. Our review highlights six recent studies exploring blockchain-based approaches to reinforce reliability, transparency, and accountability in AI-assisted SE tasks. Conclusion: Blockchain enhances trust by ensuring data immutability, model transparency, and lifecycle accountability, including federated learning with blockchain consensus and private data verification. However, inconsistent definitions of trust and limited real-world testing remain major challenges. Future work must develop measurable, reproducible trust frameworks to enable reliable, secure, and compliant AI-driven SE ecosystems, including applications involving large language models.

💡 Research Summary

The paper presents a systematic literature review (SLR) titled “Infusion of Blockchain to Establish Trustworthiness in AI Supported Software Evolution” (BAISET), aiming to map and synthesize research that combines blockchain technology with AI‑driven software engineering (SE) tools, especially those based on large language models (LLMs). The authors followed the rigorous Kitchener et al. (2017, 2018) SLR methodology, defining clear inclusion/exclusion criteria, a multi‑stage search strategy, and a quality assessment framework.

Search and Selection
Nine major digital libraries (ACM, IEEE Xplore, SpringerLink, Elsevier, MDPI, Wiley, LedgeR, TU Delft, Academia) were queried between 2017 and 2025 using carefully crafted Boolean strings. An initial pool of 603 records was reduced through title/abstract/keyword screening (Phase 1) to 75 papers, then full‑text screening (Phase 2) yielded 44 primary studies. Backward and forward snowballing added 52 more, resulting in a total of 96 studies (44 primary, 52 secondary) for synthesis.

Research Questions
Four overarching questions guided the review:

• RQ1: Motivations and methodological approaches for establishing trust with blockchain in AI contexts.
• RQ2: State‑of‑the‑art of blockchain‑enhanced trust in AI and SE.
• RQ3: Specific use of blockchain to support LLM‑based SE tools for software evolution.
• RQ4: Limitations and future research directions for BAISET.

Findings – Publication Trends
The 44 primary studies span 2017‑2025, with a peak in 2019 (10 papers). Journals dominate (56.8 % of primary studies), followed by conferences (22.7 %). IEEE and ACM venues are most frequent, and 61 % of the works appear in CORE A*/A or Q1/Q2 ranked outlets, indicating a relatively high scholarly impact.

Findings – Research Themes
Two dominant domains emerged: (1) Blockchain‑oriented SE (e.g., smart‑contract driven CI/CD, immutable artifact registries) and (2) AI‑enhanced blockchain systems (e.g., federated learning with consensus). Emerging topics include blockchain‑supported IoT, robotics, and embedded systems. However, only 31 % of the studies explicitly define “trustworthiness,” and definitions vary widely (transparency, accountability, security, auditability).

Blockchain Mechanisms for Trust
The review identifies several concrete mechanisms:

• Immutable Model/Data Hashing: Storing cryptographic hashes of training data, model weights, or generated code on-chain to prevent tampering.
• Smart‑Contract Governance: Encoding validation rules, access control, and incentive mechanisms (token rewards) for contributors in federated learning or code review pipelines.
• Audit Trails: Linking CI/CD events, test results, and LLM prompt‑response pairs to blockchain entries, enabling post‑mortem forensic analysis.
• Consensus‑Based Model Updates: Using Hyperledger Fabric or consortium chains to achieve agreement on model parameter updates, thereby ensuring that no single party can unilaterally bias the model.

LLM‑Specific Applications
Only a handful of papers address LLM4SE directly. Proposed architectures store LLM‑generated code snippets’ hashes on-chain, attach smart‑contract‑mediated review outcomes, and compute a “trust score” that can be consumed by downstream tools. Some works explore federated fine‑tuning of LLMs where each participant’s contribution is recorded on a private ledger, preserving data privacy while guaranteeing provenance.

Limitations Identified
The authors highlight three major gaps:

1. Conceptual Inconsistency – Lack of a unified, measurable definition of trustworthiness hampers cross‑study comparison.
2. Empirical Weakness – Most evaluations are limited to simulations or small‑scale prototypes; real‑world deployments in industrial settings are scarce.
3. Performance Overheads – Public blockchain transaction latency and fees can be prohibitive for high‑frequency CI/CD pipelines; scalability of consensus mechanisms remains an open issue.

Future Research Directions
The paper calls for:

• Development of standardized trust metrics and benchmark datasets for blockchain‑AI SE.
• Exploration of layer‑2 or permissioned blockchain solutions to mitigate latency and cost while preserving immutability.
• Integration of regulatory and compliance considerations (e.g., GDPR, ISO/IEC 42010) into blockchain‑AI governance frameworks.
• Larger‑scale case studies involving enterprise CI/CD environments, multi‑organization federated LLM training, and longitudinal audits of blockchain‑recorded artifacts.

Conclusion
Blockchain offers compelling technical affordances—immutability, decentralized verification, and programmable governance—that can address key trust challenges in AI‑augmented software evolution, particularly for LLM‑driven code generation and model maintenance. Nevertheless, the current literature is still at a proof‑of‑concept stage, with fragmented definitions of trust and limited real‑world validation. Advancing the field will require rigorous, metric‑driven studies, performance‑optimized blockchain architectures, and comprehensive governance models that align with both engineering practice and regulatory mandates.