Open spec for AI agent decision logging aims to boost audit transparency
AATF proposes open standard for recording AI agent reasoning to enable audit trails and explainability—critical for accounting firms deploying autonomous AI systems.
The Wire — Page 6
1 more from HN: Accounting & AI
Audit checklists for AI coding agents proposed across languages
Software engineer proposes 30-item audit checklist for validating AI-generated code across programming languages, addressing quality and safety concerns in agent-driven development.
Agentic AI offers controlled automation path for SMB accounting workflows
arXiv paper argues agentic AI can reduce administrative burden and accelerate routine processes for small-to-medium companies through multi-step task planning and enterprise system integration.
15 more from arXiv: AI + Accounting
Verifiable data-analytic agents improve auditability of LLM reasoning
Researchers propose VeriGraph, a framework to make LLM-based agents' outputs auditable by separating deterministic data computations from semantic reasoning, addressing a key challenge for accounti...
Researchers propose risk-pricing model for autonomous AI agents in business o...
Academic paper quantifies autonomous AI agent failure risk through trace-level underwriting to enable profitable, insurable deployment in operational systems including accounting workflows.
Mojo language targets reproducibility gap in quantitative finance systems
Mojo, Modular's Python-like systems language, addresses the 30-year 'two-language tax' in quantitative finance where Python research models are rewritten in C++ for production, introducing numerica...
Green SARC framework controls costs and emissions in agentic AI systems
Researchers propose Green SARC, a governance framework that enforces financial and environmental cost controls within agentic AI loops before execution, addressing runaway spending in multi-agent s...
New Framework Secures AI Agents Controlling Production Accounting Systems
Researchers propose Sovereign Assurance Boundary (SAB), a certificate-based control system for AI agents managing high-stakes financial infrastructure, addressing audit and authorization gaps in no...
Trace2Policy: AI agents learn audit rules from expert behavior traces
Researchers introduce Trace2Policy, an AI system that extracts tacit decision rules from audit and compliance experts, then iteratively refines them through error analysis—automating expert judgmen...
DeepSeek-R1 Model Trained for Financial Entity Recognition in Reports
Researchers fine-tune DeepSeek-R1-8B with LoRA and NEFTune to improve financial named-entity recognition, enabling better extraction of structured data from unstructured reports and news for knowle...
Research on multi-agent protocols for operational AI in regulated workflows
ArXiv paper on Collaborative Human-Agent Protocol (CHAP) for managing multi-human, multi-agent systems in production—relevant to accounting firms deploying AI agents across audit, tax, and complian...
Researchers tackle silent scope omission in policy-following AI agents
Academic paper proposes deontic trees to help AI agents correctly parse nested exceptions in regulations, addressing failures where systems appear compliant but miss critical edge cases in tax, aud...
DuMate-DeepResearch system tackles auditable multi-agent reasoning for comple...
New multi-agent framework DuMate-DeepResearch addresses hallucination and verification challenges in deep research tasks through recursive search and rubric-grounded reasoning, with direct applicat...
Researchers develop LoRA-MINT to audit training data in fine-tuned LLMs
New methodology LoRA-MINT enables auditing of training data membership in domain-adapted LLMs, addressing IP and data sensitivity concerns for accounting applications.
Research team proposes evidence tracing for LLM agent auditing
Academic paper introduces methods to trace and verify LLM agent decision-making through execution provenance, addressing audit and compliance needs for autonomous AI systems in high-stakes domains.
Provably Auditable LLM Agents Use Typed Lambda Calculus for Guaranteed Correc...
Agentic Redux, a new LLM agent architecture, uses typed lambda calculus to guarantee semantic correctness with append-only ledger auditability, demonstrated in healthcare billing compliance and sec...
New benchmark tests AI agents on auditable financial research workflows
Researchers introduce BigFinanceBench, a 928-item benchmark designed to evaluate financial-research AI agents on auditability and workflow transparency—measuring not just final answers but the deri...
Knowledge-Graph Pattern Brings Persistent Audit Trails to Compliance Detection
Academic paper proposes Violation Situation Pattern (VSP), a knowledge-graph approach that converts transient compliance violations into persistent audit objects with lifecycle tracking, review sta...
Tax practitioners double AI research adoption, spurring alternative billing m...
Tax professionals' AI usage for research nearly doubled year-over-year, prompting exploration of new billing approaches beyond traditional hourly rates.
AI adoption among tax firms nearly doubles in one year, survey finds
Blue J and CPA.com survey shows tax practitioners moving beyond experimentation to embed AI as core workflow tool, signaling shift from pilot projects to firm-wide transformation.