The Accountability Paradox: How Platform API Restrictions Undermine AI Transparency Mandates
Florian A.D. Burnat, Brittany I. Davidson · 2025
Read The Accountability Paradox: How Platform API Restrictions Undermine AI Transparency Mandates by Florian A.D. Burnat, Brittany I. Davidson (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-11577.txt.
Introduction
Florian Burnat and Brittany Davidson expose a clash, in a 2025 preprint, between platform rules and a European transparency law. The Digital Services Act requires platforms to give researchers data access. That access lets outsiders audit how moderation and ranking systems work. At the same time, big platforms have tightened their programming interfaces, the pipes researchers use. The authors build an audit framework to measure the gap. They compare X, Reddit, TikTok, and Meta against what the law demands.
Their map shows blind spots where key systems cannot be checked at all. The paradox is sharp: the more platforms lean on AI, the less outsiders can inspect it. The authors propose fixes guided by a national AI risk framework. Shared access models and stronger enforcement lead the list. Without them, they argue, transparency mandates will stay words on paper.
Catalog record
| CREATOR | Florian A.D. Burnat, Brittany I. Davidson |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-11577
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
The Accountability Paradox: How Platform API Restrictions Undermine AI Transparency Mandates Florian A.D. Burnat, Brittany I. Davidson Abstract Recent application programming interface (API) restrictions on major social media platforms challenge compliance with the EU Digital Services Act ( European-Parliament-and-Council2022-eg ) , which mandates data access for algorithmic transparency. We develop a structured audit framework to assess the growing misalignment between regulatory requirements and platform implementations. Our comparative analysis of X/Twitter, Reddit, TikTok, and Meta identifies critical “audit blind-spots” where platform content moderation and algorithmic amplification remain inaccessible to independent verification. Our findings reveal an “accountability paradox”: as platforms increasingly rely on AI systems, they simultaneously restrict the capacity for independent oversight. We propose targeted policy interventions aligned with the AI Risk Management Framework of the National Institute of Standards and Technology ( NIST2023 ) , emphasizing federated access models and enhanced regulatory enforcement. 1 Introduction Social media platforms have become integral to public discourse, and are critical sources of training data for contemporary AI systems. Recent restrictions on application programming interfaces (APIs) have significantly reduced the volume and granularity of data available for external examination, undermining the long-established norms of transparency ( Bruns2019-dk ; Tromble2021-ca ) . Researchers increasingly characterize the post-API environment as a “data abyss,” wherein even fundamental replication studies are no longer feasible ( de-Vreese2023-xw ) . These API restrictions have profound implications for the development and governance of AI. Large language models and recommendation systems trained on platform data inherit biases and patterns present in the data; however, researchers face increasing barriers to studying these patterns. This creates an alarming scenario in which AI systems with significant societal impacts simultaneously become more powerful and less transparent. Our analysis focuses on the European Union Digital Services Act (DSA), which establishes data access rights for “vetted researchers” ( European-Parliament-and-Council2022-eg ) . This presents a multifaceted compliance challenge for global platforms, as parallel yet divergent regulatory frameworks compel companies to satisfy inconsistent data access requirements across jurisdictions ( de-Vreese2023-xw ) . These implications extend beyond academic interests. Independent data access has facilitated the revelation of the Cambridge Analytica scandal ( Graham-Harrison2018-yh ) , the assessment of algorithmic amplification of partisan content ( Huszar2022-nw ) , and the examination of automated content moderation practices ( Roberts2021-ud ) . Without such access, potentially harmful AI behaviors, from biased content recommendations to manipulated information flows to discriminatory content moderation, may propagate undetected. The field of AI safety research, which critically depends on understanding how AI systems behave “in the wild,” is particularly hampered by these restrictions. The emerging AI alignment and evaluation methods emphasize the importance of real-world behavioral testing and independent red teaming. However, API restrictions effectively limit the ability to conduct such evaluations, concentrating power on the hands of platform operators and their selected commercial partners. This creates a dangerous gap in our collective ability to ensure that AI systems operate as intended and align themselves with societal values. To address these challenges, we propose policy interventions aligned with the AI Risk Management Framework of the National Institute of Standards and Technology ( NIST2023 ) , which emphasizes transparency and independent oversight as essential components of responsible AI governance. Our findings highlight the urgency of resolving these transparency gaps before the next generation of AI systems further entrench data asymmetries. 2 Related Work Our research builds upon the literature on platform data access, regulatory frameworks, and AI governance. We situate our contributions within four key research domains: 2.1 API Access and Platform Governance Research has documented increasing restrictions of social media APIs, a phenomenon ( Bruns2019-dk ) termed the “API-calypse.” This trend emerged after high-profile scandals, such as Cambridge Analytica ( Graham-Harrison2018-yh ; Hinds2020-up ) , and gained momentum post-2018. ( Freelon2018-tq ) described this as a shift towards a “post-API age” in computational social science, while ( Tromble2021-ca ) examined methodological implications for academic research. Recent studies by ( Davidson2023-xz ) established that these restrictions threaten principles of open science, creating what ( de-Vreese2023-xw ) characterize as a “data abyss.” Alternative methodologies have emerged in response, with ( Perriam2020-ti ) investigating digital methods within the post-API environment and others suggesting web scraping as an imperfect alternative ( Mancosu2020-by ; Harrell2024-tx ) . However, these approaches introduce significant methodological limitations ( Landers2016-cs ) . 2.2 Regulatory Frameworks for Digital Platforms Our analysis concentrates on the DSA, drawing upon ( Li2023-pj ) ’s critique of transparency obligations and ( Busuioc2022-or ) ’s examination of the logic of secrecy within the EU AI Act. ( Dommett2022-yo ) investigated challenges faced by academics advocating for platform data access, while ( Valkenburg2025-zj ) emphasized the necessity of ensuring research access to platform data. 2.3 AI Transparency and Auditing The transparency mandates that we examine connect to the broader discourse on AI system auditing. ( Raji2020-mk ) proposed a framework for algorithmic auditing, while ( Balasubramaniam2023-jh ) traced the progression from ethical guidelines to specific transparency requirements for AI systems. ( Haresamudram2023-pf ) delineated distinct levels of AI transparency, which we apply in our analysis of platform compliance. Recent advances in AI model evaluation have highlighted the importance of behavioral testing and external auditing. Bommasani2024-sk emphasize that transparency reports for foundation models require access to interaction data at scale, which current API restrictions significantly impede. Similarly, Longpre2024-es argue for a “safe harbor” for AI evaluation, an objective undermined by restricted access to platforms where these AI systems operate. We also draw upon established governance frameworks, particularly the NIST AI Risk Management Framework ( NIST2023 ) , and ( Sandvig2014-uc ) ’s pioneering work on auditing algorithms. 2.4 Ethical Tensions in Platform Research Our exploration of ethical tensions is grounded in ( Nissenbaum2009-pi ) ’s contextual integrity framework, recently applied to Reddit research by ( Fiesler2024-sl ) . ( Taylor2018-qt ) investigated approaches to ethical challenges in mining social media data, while ( Zimmer2010-yv ) highlighted tensions between open data and privacy expectations. This study advances these research streams by specifically investigating the interaction between recent API restrictions and regulatory transparency mandates, analyzing their intersection, and proposing targeted interventions to address the resulting accountability challenges in AI governance. 3 Background: API Governance and Regulatory Context Social media APIs initially promoted openness but transitioned to restrictive access following data scandals ( Graham-Harrison2018-yh ) . Since 2018, platforms have implemented graduated access tiers, stricter rate limits, and direct fees for previously free endpoints ( Bruns2019-dk ) . Notable examples include X/Twitter’s 2023 discontinuation of free tiers and Reddit’s introduction of usage-based tariffs ( Gerken2023-kq ) . European legislators have sought to re-establish research access through the DSA, which mandates data provision to “vetted researchers” investigating systemic risks ( Valkenburg2025-zj ) . Concurrently, AI governance frameworks such as the EU AI Act ( European-Parliament-and-Council2024-xa ) and NIST RMF 1.0 ( NIST2023 ) require demonstrable provenance and auditability of training data. These expectations conflict with the increasing inaccessibility of social media data for independent examination, creating an “open-science tension” precisely when downstream transparency requirements are increasing. 4 Methods 4.1 Audit Design Our methodology builds on the API restrictions documented by ( Davidson2023-xz ) and develops a structured framework to examine the intersection of these restrictions and regulatory transparency requirements. We