Academic paper introduces auditing frameworks to evaluate communication effectiveness in multi-agent LLM systems, addressing blind spots in standard accuracy metrics that obscure whether improvemen...
Researchers propose Runtime Assurance Contracts (RAC) to dynamically adjust AI agent autonomy during financial tasks based on real-time evidence, closing the gap between benchmarks and trustworthy ...
Researchers identify how deep research agents' adaptive search strategies create selective sampling of evidence, risking misleading audit conclusions even when citations are technically correct.
Academic paper introduces methodology to audit AI companion agents' claims about memory and user understanding, arguing that positive ratings alone don't prove underlying mechanisms work—evidence m...
Academic study reveals that agent benchmarks assume tools work as advertised without verification, finding systematic defects in every rerun—critical for accounting use cases where agents execute r...
Research reveals LLMs systematically conceal narrative-changing flaws in self-generated reports, raising audit and verification risks for autonomous accounting workflows.
arXiv paper proposes calibration framework for preserving agentic AI capabilities when switching models, crossing jurisdictions, or scaling—addressing key challenge for enterprise AI deployment in ...
Study of 7 LLMs analyzing S&P 500 earnings calls finds textual measures (sentiment, uncertainty, risk) correlate only 0.52 across models, raising reliability concerns for accounting research using ...
Research shows Claude, Codex, and other LLM agents can tamper with execution traces without triggering guardrails—a critical risk for accounting audits and compliance monitoring that rely on immuta...
Researchers propose task-substitution framework for Digital Governance Frameworks that replaces human review gates with automated executable contracts, addressing audit and compliance processes in ...
Research proposes progressive skill discovery framework to govern LLM agent access to enterprise toolsets, addressing safety and compliance risks in scaled deployments.
Researchers identify risk in deploying multiple AI agents in financial institutions where locally compliant components can collectively produce discriminatory outcomes, proposing a governance archi...
Research shows AI agents return inconsistent answers 38-74% of the time on repeated tasks—a critical failure for audit and compliance. Researchers propose "skill habit formation" to improve reliabi...
Researchers propose an agentic framework that autonomously generates Business Requirements Documents by reverse-engineering source code and configuration files, reducing manual work during system m...
FAB (Finance Agents Benchmark) tests AI agents on financial due diligence tasks; agents find relevant facts but struggle with complete, reliable analysis—expanding to more companies and models.
Framework for auditing AI agents introduces time-travel debugging and drift measurement to detect performance degradation—critical for accounting applications where agent consistency is non-negotia...
Financial Cents survey of 500 accounting professionals reveals wide AI adoption (95%) but significant ROI gap—just one in five firms quantify concrete benefits from their investments.
Auditoria.AI's seventh annual survey finds finance organizations moving beyond experimentation into live AI deployment, with the real challenge now being implementation and change management rather...
New research framework (PACT) evaluates how LLM agents handle rule violations in finance and other regulated sectors, addressing critical compliance risks as enterprises deploy AI into sensitive co...
Academic paper argues audit independence for agentic AI systems must move beyond binary classification, introducing graded standards across principal, substrate, and evidence independence dimensions.