While mechanistic interpretability (MI) has produced important insights into neural network internals, the field has yet to establish a standardized system to audit experiments. As such, many of its findings remain underutilized in safety-critical applications such as medical AI and autonomous systems, as stakeholders cannot certify their validity. Recent work demonstrates this concretely: two papers found conflicting conclusions for the same behavior, and a third study revealed that both were partially correct but incomparable due to methodological inconsistencies. Without standardized auditing, such ambiguities hinder adoption in high-stakes contexts requiring strong correctness guarantees. We call for the MI community to work towards developing a novel reviewing system that complements peer review via: (1) Continuous reviewing supported by a Collaborative Reviewing Platform where meta-science results and discussions (such as critiques, negative results, post-hoc extensions, reproductions, replications, and partial results) that fit outside of papers are organized and discussed, allowing for comments and revisions to be made at any time (2) Generalizing good practices found on this platform into expert-verified guidelines and protocols to improve auditing efficiency, and (3) Source-based auditing systems that track arguments which claims depend on. This position paper encourages constructive debate over the necessity, design and implementation of such a framework, providing early concrete examples to help catalyze these dialogues. Overall, we propose that auditing MI itself is essential for its application in AI safety, industry, and governance.

Mechanistic interpretability research faces a gap between analyzing simple circuits in toy tasks and discovering features in large models.To bridge this gap, we propose text-to-SQL generation as an ideal task to study, as it combines the formal structure of toy tasks with realworld complexity.We introduce TinySQL, a synthetic dataset, progressing from basic to advanced SQL operations, and train models ranging from 33M to 1B parameters to establish a comprehensive testbed for interpretability.We apply multiple complementary interpretability techniques, including Edge Attribution Patching and Sparse Autoencoders, to identify minimal circuits and components supporting SQL generation.We compare circuits for different SQL subskills, evaluating their minimality, reliability, and identifiability.Finally, we conduct a layerwise logit lens analysis to reveal how models compose SQL queries across layers: from intent recognition to schema resolution to structured generation.Our work provides a robust framework for probing and comparing interpretability methods in a structured, progressively complex setting.

Recent proposals for regulating frontier AI models have sparked concerns about the cost of safety regulation, and most such regulations have been shelved due to the safety-innovation tradeoff. This paper argues for an alternative regulatory approach that ensures AI safety while actively \textit{promoting} innovation: mandating that large AI laboratories release small, openly accessible "analog models"—scaled-down versions trained similarly to and distilled from their largest proprietary models.Analog models serve as public proxies, allowing broad participation in safety verification, interpretability research, and algorithmic transparency without forcing labs to disclose their full-scale models. Recent research demonstrates that safety and interpretability methods developed using these smaller models generalize effectively to frontier-scale systems. By enabling the wider research community to directly investigate and innovate upon accessible analogs, our policy substantially reduces the regulatory burden and accelerates safety advancements.This mandate promises minimal additional costs, leveraging reusable resources like data and infrastructure, while significantly contributing to the public good. Our hope is not only that this policy be adopted, but that it illustrates a broader principle supporting fundamental research in machine learning: deeper understanding of models relaxes the safety-innovation tradeoff and lets us have more of both.

Understanding the inner workings of machine learning models like Transformers is vital for their safe and ethical use. This paper provides a comprehensive analysis of a one-layer Transformer model trained to perform n-digit integer addition. Our findings suggests that the model dissects the task into parallel streams dedicated to individual digits, employing varied algorithms tailored to different positions within the digits. Furthermore, we identify a rare scenario characterized by high loss, which we explain. By thoroughly elucidating the model’s algorithm, we provide new insights into its functioning. These findings are validated through rigorous testing and mathematical modeling, thereby contributing to the broader fields of model understanding and interpretability. Our approach opens the door for analyzing more complex tasks and multi-layer Transformer models.

Artificial intelligence (AI) has a multitude of applications in cancer research and oncology. However, the training of AI systems is impeded by the limited availability of large datasets due to data protection requirements and other regulatory obstacles. Federated and swarm learning represent possible solutions to this problem by collaboratively training AI models while avoiding data transfer. However, in these decentralized methods, weight updates are still transferred to the aggregation server for merging the models. This leaves the possibility for a breach of data privacy, for example by model inversion or membership inference attacks by untrusted servers. Somewhat-homomorphically-encrypted federated learning (SHEFL) is a solution to this problem because only encrypted weights are transferred, and model updates are performed in the encrypted space. Here, we demonstrate the first successful implementation of SHEFL in a range of clinically relevant tasks in cancer image analysis on multicentric datasets in radiology and histopathology. We show that SHEFL enables the training of AI models which outperform locally trained models and perform on par with models which are centrally trained. In the future, SHEFL can enable multiple institutions to co-train AI models without forsaking data governance and without ever transmitting any decryptable data to untrusted servers.

This article describes the design and evaluation of two generations of an interface for navigating datasets of gigapixel images that pathologists use to diagnose cancer. The interface design is innovative because users panned with an overview:detail view scale difference that was up to 57 times larger than established guidelines, and 1 million pixel “thumbnail” overviews that leveraged the real estate of high-resolution workstation displays. The research involved experts performing real work (pathologists diagnosing cancer), using datasets that were up to 3,150 times larger than those used in previous studies that involved navigating images. The evaluation provides evidence about the effectiveness of the interfaces and characterizes how experts navigate gigapixel images when performing real work. Similar interfaces could be adopted in applications that use other types of high-resolution images (e.g., remote sensing or high-throughput microscopy).

In recent years, a significant number of developments across a broad range of disciplines have allowed researchers and clinicians to start to build up a picture of cancer development. In this paper we report upon the development of a prototype of a secure distributed infrastructure that links imaging data from pathology and radiology. The intention is that a fully-developed system will be capable of supporting studies that will examine whether prognostic and diagnostic features which are apparent in histopathological sections and clinical scans are related. Further, these studies will consider whether these features can be meaningfully linked into a diagnostic or predictive profile. The project in which the prototype is being developed naturally involves a large degree of cooperation across various disciplines. The focus of this paper is primarily on the development of the underlying prototype infrastructure.

BACKGROUND: Tissue microarray (TMA) technology has been developed to facilitate large, genome-scale molecular pathology studies. This technique provides a high-throughput method for analyzing a large cohort of clinical specimens in a single experiment thereby permitting the parallel analysis of molecular alterations (at the DNA, RNA, or protein level) in thousands of tissue specimens. As a vast quantity of data can be generated in a single TMA experiment a systematic approach is required for the storage and analysis of such data. DESCRIPTION: To analyse TMA output a relational database (known as TmaDB) has been developed to collate all aspects of information relating to TMAs. These data include the TMA construction protocol, experimental protocol and results from the various immunocytological and histochemical staining experiments including the scanned images for each of the TMA cores. Furthermore the database contains pathological information associated with each of the specimens on the TMA slide, the location of the various TMAs and the individual specimen blocks (from which cores were taken) in the laboratory and their current status i.e. if they can be sectioned into further slides or if they are exhausted. TmaDB has been designed to incorporate and extend many of the published common data elements and the XML format for TMA experiments and is therefore compatible with the TMA data exchange specifications developed by the Association for Pathology Informatics community. Finally the design of the database is made flexible such that TMA experiments from several types of cancer can be stored in a single database, which incorporates the national minimum data set required for pathology reports supported by the Royal College of Pathologists (UK). CONCLUSION: TmaDB will provide a comprehensive repository for TMA data such that a large number of results from the numerous immunostaining experiments can be efficiently compared for each of the TMA cores. This will allow a systematic, large-scale comparison of tumour samples to facilitate the identification of gene products of clinical importance such as therapeutic or prognostic markers. In addition this work will contribute to the establishment of a standard for reporting TMA data analogous to MIAME in the description of microarray data.