FinTrust/FAB: Trust Frameworks in Finance

• FinTrust/FAB is a family of frameworks that enhance financial trust via blockchain auditing, authenticated BFT consensus, zero trust architectures, AI benchmarks, and encrypted analytics acceleration.
• It integrates cryptographic mechanisms, automated workflows, and adaptive trust scoring to meet stringent security, regulatory, and performance standards.
• Empirical evaluations show improved transaction verification speed, consensus efficiency, and scalable privacy-preserving analytics across financial applications.

FinTrust/FAB denotes a family of frameworks, protocols, and benchmarks advancing trust, automation, and performance in financial systems. They address distinct but complementary domains: (1) blockchain-based auditing automation in public markets (“FutureAB/FAB”) (Cao et al., 2020), (2) robust authenticated BFT consensus (“FinTrust/FAB protocol”) (Abraham et al., 2021), (3) financially-aware zero trust architectures (e.g., SecureBank) (Biao, 29 Dec 2025), (4) financial trustworthiness benchmarks for AI models (FinTrust LLM Benchmark) (Hu et al., 17 Oct 2025), and (5) hardware acceleration for encrypted analytics (FAB FHE Accelerator) (Agrawal et al., 2022). This entry surveys core definitions, protocol mechanisms, system architectures, evaluation methodologies, and empirical results across these, emphasizing rigorous alignment with regulatory, security, and performance requirements in finance.

1. Blockchain-Based Auditing and Trust Automation in Public Markets

FutureAB (Future Auditing Blockchain, sometimes denoted "FAB") implements a multi-layer blockchain architecture to automate transaction verification and audit processes among public firms, auditors, and regulators (Cao et al., 2020). Its key elements include:

• Architecture: Three-layer design comprising ABWallet (client-side cryptographic wallet and key management module), a web/smart-contract layer (permissioned web front-end interfacing with Ethereum smart contracts), and a consortium permissioned distributed ledger (Ethereum chain restricted to auditors, participants, and regulators).
• Cryptographic Foundations:
  • Modified Pedersen-type commitment schemes ($C = g^m h^r \mod p$) secure optional transaction details, affording both hiding and binding properties.
  • Each message is authenticated with ECDSA signatures, supporting non-repudiation and audits.
• Automated Workflow: Transactions are double-posted (sender and receiver sides) with the smart contract engine automatically matching paired entries. Discrepancies trigger "RiskFlagged" events, prompting audit intervention. Unmatched entries are flagged as "Pending." Auditors respond by requesting opening of commitments and can utilize graph-based anomaly detection over the on-chain transactional graph.
• Performance: Transaction encryption (commit+sign) measured at 0.021 s/txn; on-chain verification 0.001 s/txn. Bulk performance enables verifying 10,000 transactions in under 1 min.
• Trust Mechanisms: Blockchain immutability, complete on-chain transparency for verified transactions, multi-party auditability, and quantitative trust metrics—such as matching rate, alert rate, and verification latency—collectively enhance investor and regulatory trust.

2. Fast Authenticated Byzantine (FAB) Consensus Protocol

The FinTrust/FAB protocol presents a 2-round authenticated BFT state machine replication (SMR) protocol for permissioned blockchain, reducing commit latency and lowering replica threshold relative to prior approaches (Abraham et al., 2021):

• System Model:
  • $n$ replicas tolerate up to $f$ Byzantine faults, requiring only $n \geq 5f-1$, exceeding prior bounds (e.g., FaB's $n \geq 5f+1$).
  • Assumes partial synchrony with PKI-based digital signatures for all messages.
• Protocol Mechanics:
  • Each round consists of a leader proposal followed by replica votes, with blocks committed upon collecting $\geq n-f$ votes (QC).
  • Timeout and view-change subprotocol ensures safe lock transfer and liveness through Timeout Certificates (TC).
• Safety and Liveness:
  • Safety arises from quorum intersection and lock-transfer lemmas that prevent double-certification of diverging blocks.
  • Liveness is guaranteed post-GST under an honest leader, with commit latencies of two message rounds in the common case.
• Performance:
  • Good-case latency of 2 rounds vs. the 3 of PBFT.
  • Fewer replicas than prior FaB and strictly improves resiliency bounds for $f=1$ (4 replicas suffices).
• Comparison Table:

Protocol	Replicas Required	Good-case Latency	Authentication
PBFT	$n \geq 3f+1$	3 rounds	MACs
FaB	$n \geq 5f+1$	2 rounds	MACs
FinTrust/FAB	$n \geq 5f-1$	2 rounds	Digital sigs

3. Financial Zero Trust and Financially-Aware Banking (FAB) Architectures

SecureBank embodies the FinTrust/FAB architecture for high-assurance banking by integrating financial semantics, risk modeling, adaptive identity, and impact-driven security automation (Biao, 29 Dec 2025):

• Foundational Definitions:
  • Financial Zero Trust (FinTrust): Security policies (decisions) are functions of composite trust score ($n$0), Financial Threat Score (FTS), and segmentation context ($n$1).
  • Financially-Aware Banking (FAB): Holistically aligns security controls with business flows (payments, settlement, AML/KYC, risk analytics), transactional semantics, and regulatory requirements (e.g., PCI-DSS, DORA).
• Core Components:
  • Multi-point enforcement at API gateway, micro-segmentation, application PDP, and data-access firewall.
  • Adaptive trust scoring:

  $n$2

  supporting exponential trust decay on anomalies. - Contextual segmentation by workflow zone, with adaptive segmentation strength quantifying defense granularity. - Automation executes responses based on FTS risk bands, orchestrating actions (block, step-up, quarantine, alert) per regulatory impact.

• Evaluation Metrics:

  • Transactional Integrity Index (TII), Identity Trust Adaptation Level (ITAL), Security Automation Efficiency (SAE). SecureBank yields $n$3 for TII, $n$4 for ITAL, and $n$5 SAE, improving trust adaptation and automation coverage versus baselines.
• Regulatory Alignment: Full compatibility with PCI-DSS 4.0, NIST CSF 2.0, DORA, NIS2, and integration with SIEM/SOAR/fraud detection platforms.

4. FinTrust Benchmark: Trustworthiness Evaluation for Financial LLMs

The FinTrust benchmark (distinct from infrastructure/protocol meanings) establishes a comprehensive, multi-dimensional benchmark for evaluating the trustworthiness of LLMs in finance (Hu et al., 17 Oct 2025):

• Scope and Design: 15,680 instances across truthfulness, safety, fairness (personal, industry), robustness, privacy, transparency, and knowledge discovery. Modalities include tabular, textual, and time-series data, supporting diverse input/output forms.
• Metrics: Dimension-specific, including correctness/informativeness for truthfulness, 1–10 violation scores for safety (across four attack vectors), stability and Cohen's $n$6 for fairness, Likert-scaled robustness, privacy refusal rates, fiduciary disclosure rates, and knowledge scores.
• Empirical Summary: Proprietary LLMs (o4-mini) outperform open-source on safety; DeepSeek-V3 sets the industry fairness baseline ($n$7), but all models lag on legally required transparency and privacy disclosures (<15% disclosure rates).
• Critical Gaps: Persistent gaps in fiduciary duty, transparency, and legal compliance, as well as vulnerabilities to jailbreak attacks and privacy leakage, underscore the need for targeted alignment.

5. FAB: FPGA-Based Accelerator for Bootstrappable Fully Homomorphic Encryption

FAB, in the context of hardware, refers to an FPGA-based hardware accelerator for bootstrappable FHE (specifically CKKS “approximate” schemes) targeting encrypted analytics in FinTrust deployments (Agrawal et al., 2022):

• Cryptosystem Implementation: CKKS over $n$8, with $n$9, $f$0 levels, $f$1. Supports full bootstrapping (depth 17) and 6 fresh levels post-bootstrapping.
• Hardware Microarchitecture:
  • 256 parallel scalar FHE units, supporting modular arithmetic and slot automorphisms.
  • Balanced 43 MB on-chip SRAM and 460 GB/s HBM2 bandwidth to avoid memory bottlenecks present in prior FPGA or ASIC designs.
  • Deep pipelining, advanced scheduling, interleaved data-fetch and compute, and multi-chip expansion via 100 GbE.
• Performance:
  • Amortized per-slot multiplication $f$2s at 300 MHz; 213× CPU, 1.5× GPU speedup. Logistic regression over encrypted data (30-iteration, eight FPGAs): 0.081 s/iteration, 456× CPU speedup.
  • Near-ASIC throughput at commodity hardware cost; scalable via cloud FPGA provisioning.
• Integration in FinTrust: Enables privacy-preserving analytics and ML within FinTrust workflows; clients upload encrypted data, FAB nodes execute on ciphertexts, and only result ciphertexts are returned for local decryption—preserving full data confidentiality at scale.

6. Synthesis: Convergence and Impact Across Finance

The FinTrust/FAB line of work spans blockchain-based audit automation, fast BFT protocols, financial zero trust architectures, AI trustworthiness benchmarks, and encrypted analytics acceleration. Core themes include:

• Maximizing system trust via cryptographically enforced immutability, verifiability, and transparency.
• Achieving business-aligned automation that reduces manual workloads while focusing human intervention on high-risk and anomalous cases.
• Rigorous alignment with regulatory and fiduciary requirements, from continuous transaction traceability to model-level disclosure.
• Quantitative, benchmark-driven evaluation methodologies that empirically validate improvements in system trust, automation efficiency, compliance, and fairness.

As a result, FinTrust/FAB frameworks and technologies collectively underpin the next generation of secure, auditable, and efficient financial infrastructure, supporting both operational robustness and regulatory trust mandates (Cao et al., 2020, Abraham et al., 2021, Biao, 29 Dec 2025, Hu et al., 17 Oct 2025, Agrawal et al., 2022).