Why traditional transparency reports can't keep up with today's AI
The core problem is that modern AI systems, especially generative AI and agentic AI, behave fundamentally differently from earlier models. Generative AI can create new content—text, images, even music—rather than just analyzing existing data, which introduces unpredictability [2]. Agentic AI goes further by perceiving inputs, reasoning, planning, and executing experiments with minimal human intervention [3]. These capabilities mean that a model's behavior can change based on its own decisions, making static transparency reports quickly outdated.
Traditional reporting checklists, like the TRIPOD-AI guidelines for medical AI, focus on dataset integrity, feature selection, and model validation [1]. While valuable, these checklists were designed for predictive models that don't generate new content or act autonomously. As a result, they miss key sources of variability introduced by generative and agentic systems, such as data lineage, model specification, and agent policies [3]. The Lancet Regional Health-Europe editorial notes that generative AI has 'surpassed previous technological revolutions' in its potential, but this leap also means existing transparency frameworks are playing catch-up [2].
New frameworks and automated tools are emerging to close the transparency gap
Researchers are developing specialized guidelines and automated systems to address the shortcomings of traditional transparency reports. The TRACE-AI framework, for example, provides a publication checklist specifically for agentic AI in catalysis science, emphasizing end-to-end traceability across the full lifecycle of an AI campaign [3]. This includes linking research objectives to data and models, agent reasoning and action, and the knowledge acquired. By promoting standardized and accountable reporting, TRACE-AI aims to build trust as autonomous AI laboratories become more common [3].
Automated annotation systems offer another promising solution. One proposed framework uses automated annotations of both data and AI models to give enterprises visibility into potential issues, prepare transparency reports, and ensure policy compliance [5]. These annotations can flag problems like data quality issues or algorithmic bias, making it easier to keep reports current as models evolve [5]. In the medical domain, the TRIPOD-AI checklist provides a structured framework for evaluating AI research, including criteria for dataset integrity, model validation, and reporting transparency [1]. While not a complete solution for generative AI, it represents a step toward more rigorous and reproducible reporting.
Explainable AI (XAI) techniques like SHAP and LIME are also being used to improve transparency. In a study on DDoS detection in SDN networks, researchers used SHAP and LIME to interpret model decisions, achieving near-perfect accuracy (0.9999) while also making the model's reasoning transparent [4]. This shows that transparency and high performance can coexist, though the approach is still being adapted for more complex generative and agentic systems.
The remaining challenges: bias, reproducibility, and adoption
Even with new frameworks, significant challenges remain. Algorithmic bias, data quality issues, and limited external validation continue to undermine the reliability of AI models [1]. For example, AI models trained on small patient cohorts with thousands of candidate features are prone to overfitting, which transparency reports must explicitly address [1]. The TRACE-AI guidelines acknowledge that risks are 'particularly pronounced' in fields like heterogeneous catalysis, where subtle variations in synthesis conditions can dramatically affect outcomes [3].
Another hurdle is adoption. Many of these frameworks are still in early stages—the TRACE-AI checklist was published in 2026 and has zero citations, while the automated annotation framework from 2021 has only four [3][5]. Widespread adoption will require buy-in from researchers, publishers, and industry. The Lancet editorial notes that generative AI's potential is 'limitless,' but realizing that potential responsibly will depend on whether transparency practices can evolve as quickly as the technology itself [2].
About These Sources
This answer is built on 5 studies (4 peer-reviewed, 1 preprint) — published from 2021 to 2026, 3 from 2024 or later, 2 in Q1 journals — selected as the most relevant from 6 studies that passed quality screening, drawn from 55 papers retrieved from a database of over 500 million.
Sources used in this answer
A practical guide for nephrologist peer reviewers: evaluating artificial intelligence and machine learning research in nephrology
This practical guide for nephrologist peer reviewers proposes a structured framework using the TRIPOD-AI checklist to evaluate AI/ML research, emphasizing dataset integrity, model validation, and reporting transparency to address challenges like algorithmic bias and overfitting.
Embracing generative AI in health care
This editorial argues that generative AI (e.g., ChatGPT, Bard) represents a technological leap beyond previous inventions, with the ability to create original content and potentially replace clinical trials, but notes that existing transparency frameworks are not designed for these capabilities.
Transparent Reporting for Agentic Catalysis Enabled by Artificial Intelligence (TRACE-AI): Community Guidelines and A Publication Checklist
The TRACE-AI guidelines provide a publication checklist for agentic AI in catalysis, emphasizing end-to-end traceability across the full lifecycle of an AI campaign to address risks to reproducibility and trust in autonomous laboratories.
Deep Learning-Based DDoS Detection in SDN Networks with Explainable AI Transparency
This study applied deep learning models (CNN, LSTM, RNN, GRU, ANN) to DDoS detection in SDN networks, achieving near-perfect accuracy (0.9999) with ANN, and used XAI techniques (SHAP, LIME) to improve transparency of model decisions.
Automated Annotations for AI Data and Model Transparency
This paper proposes a framework for automated annotations of data and AI models to enable transparency reports, policy compliance, and issue visibility, though it notes challenges and opportunities for implementation.