Blockchain, the distributed ledger technology underlying cryptocurrencies, offers a tamper-resistant record of transactions that can establish trust among mutually distrusting parties. In the context of IoV, blockchain can secure vehicle-to-vehicle communications, authenticate roadside units, and manage reputation systems for connected vehicles. Federated learning, by contrast, enables machine learning models to be trained across distributed devices without sharing raw data. Vehicles can collaboratively improve traffic prediction models, detect anomalies, and optimize network resources while keeping their local data private.

The survey systematically reviews the individual contributions of blockchain and federated learning to IoV security and privacy, then examines their integration. The authors identify major challenges that arise when combining these technologies, including the computational overhead of blockchain consensus mechanisms, communication latency in federated aggregation, and the difficulty of establishing trust in highly dynamic vehicular networks where vehicles join and leave frequently.

One key insight from the survey is that blockchain and federated learning are naturally complementary. Blockchain provides the trust infrastructure that federated learning lacks—a verifiable record of model updates, participant reputation, and aggregation results. Federated learning, in turn, brings intelligence to blockchain networks, enabling anomaly detection, malicious participant identification, and adaptive consensus mechanisms. The survey proposes a reference architecture that integrates both technologies, with blockchain managing participant registration, model update verification, and reputation tracking, while federated learning handles distributed model training across vehicles.

The authors also examine practical deployment considerations, including the choice of consensus mechanisms suited to resource-constrained vehicles, the design of incentive schemes to encourage participation, and the integration of edge computing infrastructure to reduce latency. They highlight the role of roadside units and edge servers as aggregation points for federated learning and as blockchain nodes for transaction validation.

Looking forward, the survey identifies several open research directions, including lightweight consensus mechanisms for vehicular blockchain, privacy-preserving aggregation techniques that resist poisoning attacks, adaptive trust models that respond to network dynamics, and cross-domain federated learning frameworks that enable knowledge sharing across different vehicular networks. The authors argue that realizing the full potential of IoV will require continued innovation at the intersection of distributed trust and distributed intelligence.

The distinction between blockchain-based, federated-learning-based, and integrated approaches matters because each addresses a different subset of IoV requirements. Blockchain-based solutions in the vehicular domain have historically concentrated on authentication and trust: batch authentication protocols that allow roadside units to verify groups of vehicles efficiently, key agreement schemes anchored to roadside infrastructure, and multi-channel designs that separate transaction types to reduce congestion on a single ledger. Federated-learning-based solutions, meanwhile, have grown out of the recognition that vehicles generate sensitive data—position traces, camera feeds, in-cabin sensor readings—that cannot simply be uploaded to a central server. Training intrusion detection models, driving behavior classifiers, and traffic forecasting systems across fleets keeps that data local while still pooling the statistical benefit.

The integrated BC-FL paradigm, as the survey frames it, is not a single technique but a family of design patterns. In one pattern, blockchain serves as an audit layer beneath federated learning: every model update, aggregation round, and participant decision is committed to the ledger, so that any tampering with the training process becomes detectable after the fact. In another pattern, the ledger stores participant reputation scores derived from the quality of past contributions, which then feed into the selection of clients for subsequent training rounds. A third pattern reverses the roles, using learning models to optimize blockchain operations themselves—for instance, predicting transaction load or tuning consensus parameters in response to vehicular mobility patterns.

The vehicular setting imposes constraints that distinguish it from the edge-computing scenarios where BC-FL integration was first studied. Vehicles move at high speed, producing rapidly changing network topologies in which a participant present at the start of a federated training round may be unreachable before aggregation completes. Handoffs between roadside units can interrupt communication mid-round, and the wireless channel itself is subject to interference and varying capacity. These conditions make the choice of aggregation timing, participant selection, and communication scheduling far more consequential than in static deployments. The survey’s comparative analysis reflects this, treating network dynamics not as a nuisance parameter but as a first-class design consideration that shapes which combinations of blockchain and learning techniques are viable.

Security analysis in the survey spans both layers of the integrated stack. At the blockchain layer, threats include selfish mining, eclipse attacks on lightweight vehicular nodes, and the risk that roadside units acting as validators are compromised. At the federated learning layer, poisoning attacks—where malicious vehicles submit corrupted model updates—remain among the most difficult problems, since a single poisoned update can degrade a shared model used by the entire fleet. The integration is intended to mitigate precisely this class of threats: by recording updates on a ledger and tying contributions to verifiable identities and reputations, the system gains a mechanism for attributing and excluding bad actors. But the survey is careful to note that this protection is only as strong as the reputation mechanism itself, which can itself become a target.

Privacy considerations cut across the two technologies in subtle ways. Federated learning is often assumed to be privacy-preserving because raw data never leaves the vehicle, yet model updates can leak information about local datasets through gradient inversion or membership inference attacks. Blockchain, conversely, is often assumed to be transparent to a fault—every transaction is visible—which is at odds with the confidentiality expectations of vehicle owners and manufacturers. The survey’s treatment of privacy-preserving BC-FL systems therefore emphasizes the need for additional safeguards such as differential privacy noise, secure aggregation protocols, and cryptographic commitments, and it catalogs work that combines these tools rather than treating either technology as sufficient on its own.

The reference architecture synthesized in the paper is notable for deriving its structure from recurring patterns across the literature rather than from a single proposed system. Common elements include a registration and identity layer anchored in the blockchain, a learning orchestration layer that coordinates training rounds across vehicles and edge nodes, a verification layer that validates model updates before aggregation, and an incentive layer that compensates participants for computation, bandwidth, and data quality. This layered decomposition gives researchers a shared vocabulary for comparing proposals that otherwise differ in terminology, and it exposes which layers are well-studied and which remain thin in the current literature.

Among the gaps the survey identifies, several stand out for their practical significance. Most published BC-FL schemes for IoV are evaluated in simulation with optimistic assumptions about vehicle participation and channel reliability; validation on realistic mobility traces and hardware-constrained platforms remains rare. The energy cost of running consensus on vehicles with limited battery budgets is frequently acknowledged but seldom quantified in end-to-end evaluations. Interoperability across manufacturers—where different vehicle brands would need to participate in a common federated model under a shared ledger—raises governance questions that technical protocols alone cannot answer. And the interaction between federated learning and evolving regulatory frameworks for automotive data, which differ across jurisdictions, adds a compliance dimension that the technical literature has largely deferred.

The survey also situates IoV within the broader landscape of blockchain-federated learning research in adjacent domains, including industrial internet of things, healthcare, and drone networks. Cross-domain comparisons reveal which solutions generalize and which depend on vehicular-specific assumptions such as predictable road geometry or infrastructure density. For practitioners, the value of this unified treatment lies in its taxonomies: rather than cataloging papers chronologically, the authors classify approaches by the function blockchain performs in the learning pipeline, the learning task being solved, and the network architecture assumed, allowing readers to locate the subset of prior work most relevant to a given deployment scenario.

As connected vehicles become more autonomous and more data-intensive, the stakes of getting distributed trust and distributed intelligence right grow accordingly. The survey’s argument—that neither blockchain nor federated learning alone suffices for the security, privacy, and scalability demands of IoV, but that their principled integration can—frames a research agenda likely to occupy the field for years. The open challenges it enumerates, from lightweight consensus to poisoning-resistant aggregation to adaptive trust, provide a concrete map of where the next generation of contributions is most needed.

Subject of Research: Blockchain and federated learning integration for Internet of Vehicles security

Article Title: Blockchain–federated learning integration in the internet of vehicles: analysis, taxonomies, and future directions

Article References: Cherbal, S., & Gawanmeh, A. (2026). Blockchain–federated learning integration in the internet of vehicles: analysis, taxonomies, and future directions. Cluster Computing, 29(15), Article 855. https://doi.org/10.1007/s10586-026-06637-1

Image Credits: AI Generated

DOI: 10.1007/s10586-026-06637-1

Keywords: blockchain, federated learning, Internet of Vehicles, IoV security, distributed trust, privacy-preserving machine learning, federated, learning, integration, internet, vehicles, analysis