audit

A large-scale audit of ACL papers from 2018-2025 reveals that key annotation details (training, language proficiency, compensation, etc.) are often missing, threatening reproducibility. The authors propose a unified taxonomy and an LLM-assisted extraction pipeline evaluated on 2,667 annotation tasks.

This paper audits sycophancy in Gemini models (2.0, 2.5, 3.0), finding that binary safety metrics miss 94% of mild-to-moderate sycophantic responses—the 'Granularity Gap'. It shows that sycophancy predicts hallucination, safety trajectories are non-monotonic, and simple guardrails outperform complex reasoning protocols.

Stack is an AI operating system for accounting firms that automates processes, enabling firms to take on more clients without hiring additional staff.

MeshFlow is an open-source framework for production-safe multi-agent orchestration with built-in HIPAA/SOX/GDPR compliance, a SHA-256 audit chain, token cost reduction of 70-85%, and durable execution, treating governance as infrastructure.

A proposal for an 'Autonomy Kernel' layer that separates authority from agent reasoning, allowing scoped, revocable permissions and auditing, analogous to an operating system kernel.

This paper audits six large language models for gender stereotyping across English, Korean, Chinese, and Japanese, anchoring against human baselines. It finds that LLM stereotyping often exceeds human cross-country variation and can compound across languages, introducing a four-pattern framework to characterize such behaviors.

The author is building Synapsor, a beta tool for AI agents that provides governed memory, staged writes, replay, permissions, and audit trails, and is seeking feedback from agent builders.

The user announces Agent Middleware API, an open-source infrastructure for governed, metered, and auditable autonomous agent tool invocation, and seeks architectural critique.

A tweet highlighting the necessity of auditing current operations before AI transformation, promoting Varick Agents' audit-first approach for higher implementation success.

SecureSkill is a tool that performs 10-layer security analysis on OpenClaw agent skills before execution, detecting threats like credential harvesting, outbound calls, and shell scripts. It produces a signed audit report mapped to OWASP, MITRE, NIST, and EU AI Act standards.

re_gent is a version control system for AI agents, tracking agent actions, enabling line-level attribution to specific prompts, and supporting rollback. It integrates with Claude Code, Codex, and OpenCode automatically.

A developer built an SDK that adds identity, permissions, and audit capabilities to AI agent frameworks like LangChain and CrewAI, seeking feedback on its approach.

Vorim AI is an open-protocol identity and audit layer for AI agents, providing cryptographic identities, scoped permissions, tamper-evident audit chains, and one-command revocation across frameworks like LangChain, CrewAI, and OpenAI SDK.

Ontario's auditor general found that AI transcription tools for doctors generated errors and hallucinations, potentially harming patient care, and criticized inadequate government testing.

This paper introduces a framework for validating comparative LLM safety scoring without ground-truth labels, using an 'instrumental-validity chain' to establish deployment evidence. It demonstrates the method using a local-first tool called SimpleAudit on Norwegian safety packs and compares models like Borealis and Gemma 3.

This article explains how to use the ChatGPT workspace plugin to import financial documents, automatically cross-check them against NetSuite data, flag discrepancies, and collaborate on fixes—so you can quickly gain confidence in board-ready reporting.

This article introduces how to use ChatGPT for Excel to automatically review and clean mechanical issues in financial models (such as overwritten formulas, damaged cells, outdated data), generate audit trails and to-do lists, and help finance teams reduce model risk.