E2A-Bench Tests Whether Financial Chart VLMs Turn Evidence Into Reliable Actions

A new arXiv paper introduces E2A-Bench, a benchmark aimed at measuring how reliably financial vision-language models convert chart evidence into action recommendations. The authors argue that existing hallucination tests focus on whether individual claims are supported, rather than whether the underlying evidence actually drives the recommended action. The benchmark is designed to close that gap in evaluating financial chart reasoning.

PortBench: Correlation-Aware Benchmark for LLM Portfolio Management

Researchers introduce PortBench, a benchmark for evaluating large language models on portfolio management tasks. It addresses gaps in prior benchmarks by covering multiple asset classes and accounting for cross-asset correlations across the full pipeline. The work aims to give a more realistic measure of LLM performance in financial portfolio settings.

FundaPod: Multi-Persona Agent Architecture with Knowledge Graph Memory for Investment Research

A new arXiv paper proposes FundaPod, an architecture that splits fundamental investment research work across several LLM agents, each taking a different analytical persona. The system stores findings in a knowledge graph so that information persists and can be reused across the pod rather than being lost between steps. It targets institutional-style equity research, an area the authors say has received less attention than prediction-focused financial LLM work.

Audio encoders detect managerial evasiveness in earnings calls

A new arXiv preprint presents an approach that uses conversational audio encoders to spot evasive language from managers during earnings conference calls. Rather than aggregating vocal and lexical features across an entire call, the method analyzes the conversational dynamics between analysts and executives. Prior research has tied such cues to later negative outcomes for firms, and this work aims to capture them more precisely.

Study Tests Whether LLMs Can Simulate Individual Financial Decisions

A preliminary arXiv paper examines whether large language models can stand in for people as user simulators in financial settings. The authors ran a controlled paper-trading experiment with 120 volunteers to see how well model behavior tracks evolving individual investing choices. They conclude that current LLM simulation of such decisions is not yet reliable and call for further work.

Paper Proposes Training-Free Method for Analyzing SEC Filings

A new arXiv preprint describes a method for corporate and financial-disclosure analysis that avoids training and cross-model alignment steps. The authors argue that dense text embeddings and large language models struggle with context limits, hallucination risk, compute cost, and inconsistent vector spaces across independently trained models. The work is demonstrated on SEC filings, with a revised version now posted.

CLQT benchmark targets diagnostic evaluation of LLM portfolio-management agents

A new arXiv paper introduces CLQT, a closed-loop, cost-aware and strategy-consistent benchmark for evaluating LLM agents that manage investment portfolios. The authors argue that ranking agents by returns over a fixed window fails to show whether their process is sound or their performance durable, and propose a diagnostic alternative. The work is cross-listed in cs.AI and cs.LG as a replacement submission.

FinAcumen agent uses self-evolving experience memory for financial multimodal reasoning

A new arXiv paper introduces FinAcumen, an agent design for financial tasks that must combine arithmetic, information retrieval, chart and document interpretation, and time-based reasoning over mixed evidence. While earlier tool-using agents focused mainly on executing steps accurately, this work targets the remaining difficulty of handling heterogeneous inputs, proposing a memory harness that evolves from its own accumulated experience. The submission is a revised version of the preprint.

Claude iOS app hints at "Money" feature with bank account access

A new section discovered in the Claude iOS app suggests Anthropic may be preparing a financial tool called "Money." According to t3n, the feature could give the assistant direct access to users' bank accounts, though details remain limited. Anthropic has not announced the feature officially, so it is still unconfirmed.

arXiv paper proposes data-centric post-training pipeline for financial reasoning

A new research paper tackles the shortage of training data suitable for reasoning-focused fine-tuning in the financial domain, noting that most available QA pairs lack explicit reasoning steps, sufficient context, or reliably checkable answers. The authors present a pipeline that mines financial text, distills it into reasoning-oriented training examples, and applies learning with verifiable answers to improve model performance on financial tasks.

arXiv paper integrates prediction with multi-period portfolio optimization

A machine learning paper posted on arXiv (2512.11273, updated to v3) presents an integrated approach combining prediction with multi-period portfolio optimization. The framework accounts for transaction costs, path-dependent risks, and the sequential structure of trading decisions that single-period models fail to capture. The work targets practical portfolio management where trades unfold over multiple periods.

Study finds generative AI reshapes financial analysts' information production

A working paper examines the 2023 integration of generative AI into the FactSet platform, treating the rollout as a natural experiment in AI access for finance professionals. The authors report that the technology measurably changed how equity analysts produce research and process information. The findings offer early evidence on how generative AI affects knowledge work in the financial industry.

Study tests whether financial sentiment tools validate the same way for labels and market signals

A new arXiv paper examines a common assumption in financial NLP: that a sentiment model validated against human annotations can be trusted to extract market signals. The authors measure both objectives in a setting where they can be compared directly, finding that the two evaluations do not necessarily capture the same thing. The work suggests sentiment tools may need separate validation depending on whether they are used for labeling text or predicting prices.

Context-Augmented LLMs Used to Improve Financial Forecasting with Alternative Data

A new arXiv paper examines how large language models can incorporate alternative data sources, such as consumer transactions, web traffic, and prediction markets, when forecasting a company's future financial performance. The authors argue these non-traditional signals offer timely insight into a firm's operating activity and propose augmenting LLMs with contextual information to make such data usable in forecasting tasks.

Paper Examines How Adversarial Trading Behavior Spreads Across Agents

A new arXiv paper argues that checking individual transactions is insufficient for spotting manipulation, because adversarial market behavior can be spread across multiple messages, agents, assets, and time. The authors study this interpretation gap using a virtual exchange populated by ten role-conditioned agents, aiming to move beyond transaction-level controls toward market-wide analysis. The work is relevant to those building or auditing autonomous financial agents.

French regulator ACPR says EU AI Act and DORA fall short for frontier models

Denis Beau, the incoming head of France's ACPR banking and insurance regulator, argued that existing EU rules such as the AI Act and DORA are not adequate for advanced frontier models. He proposed a staged rollout, limiting access to trusted G7 partners and building European evaluation capacity. The stance matters because nearly all French banks and insurers are already exposed to these technologies.

OpenAI Launches ChatGPT for Financial Services With Built-In Market Data

OpenAI has introduced a version of ChatGPT tailored to financial services, pairing built-in financial data with its GPT-6 Astra model. The offering is aimed at research, modeling, and producing client-ready materials. It marks a further push by the company into industry-specific enterprise products.

WHY IT MATTERS ↘Bundling licensed market data into a vertical model shifts competition from raw model capability to data rights and compliance, making generic LLM wrappers in regulated finance harder to defend. It also signals that frontier labs will increasingly monetize through proprietary data partnerships and audit-ready enterprise features rather than API access alone.