[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83823-en":3,"doc-seo-83823-105":29,"detail-sidebar-cat-0-en-105":90},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},83823,2336464648322,"Aria","https://ap-avatar.wpscdn.com/avatar/2200025388227c56fec?_k=1778556882303663488",8,"Research & Report","Government AI Use as a Monitoring Primitive A Public Document Pilot Study","Frontier AI governance depends on observing how governments adopt and use AI systems, yet direct measurement of day-to-day practice is difficult. A complementary monitoring primitive measures traces of language-model assistance in public government documents. The method is lightweight, externally reproducible, and grounded in revealed behavior rather than stated intent. A pilot study of ten U.S. and PRC-related document streams finds near-zero signals in 2021, rising by 2026, with different downstream concentrations by country. The work evaluates how this signal can complement existing monitoring tools and where it falls short.","Government AI Use as a Monitoring Primitive: A Public Document Pilot Study  \nDavid I. Atkinson 1 Joan Eleanor O’Bryan 2  \narXiv :2607 .04543v 1 [ cs .CY] 5 Jul 2026  \nAbstract  \nGovernments are important actors in frontier AI governance, but many facts about their adoption and use of AI systems are difficult to observe directly. Procurement disclosures and official statements are useful, but can also be delayed, selective, and better suited to measuring formal adoption than actual day-to-day use. We propose a complementary monitoring primitive: measuring traces of language-model assistance in public government documents. The approach is lightweight, externally reproducible, and based on revealed behavior rather than stated intent. In a pilot study of ten public document streams from U.S. and PRC government-related sources, we find that, while 2021 baselines are consistently near zero, by 2026, four of our ten sources show statistically significant signs of AI-assisted writing. In our sample, the U.S. signal concentrates in publications downstream of policy work; the PRC signal concentrates closer to it. We close by discussing how this signal could complement existing instruments for monitoring government AI adoption, and where it falls short.  \nIn 1954, the economist Armen Alchian famously used publicly traded equity prices to infer that lithium was the fusion fuel in the newly developed hydrogen bomb (Newhard, 2014) . In doing so, he neatly illustrated a more general principle: strategically important facts that are not directly observable often leave indirect traces in public data.  \nOne of these strategically important facts is the degree to which governments are familiar with, and using, frontier AI models. Will governments use frontier AI systems for analysis, propaganda, military planning, cyber operations, or regulatory enforcement? Will they restrict their own use, subsidize deployment, or quietly integrate models into routine bureaucratic work? Better answers to these ques-  \n1Northeastern University 2Harvard University. Correspondence to: David Atkinson \u003C[atkinson.da@northeastern.edu](atkinson.da@northeastern.edu) > .  \nSecond Workshop on Technical AI Governance Research (TAIGR)@ ICML 2026, Seoul, South Korea. 2026. Copyright 2026 by the author(s) .  \nMean A I Fraction  \n0.10  \n0.05  \n0.00  \n\n|  China |  |\n| --- | --- |\n|  United States |  |\n| \u003Cbr> |  |\n|  |  |\n\n2021 2024 2025 2026  \nYear  \nFigure 1. Pooled AI-writing signal by country, 2021–2026 . Each marker is the mean per-document fraction of text flagged as AIgenerated by Pangram (range 0–1), pooled across each country’s sources (Figure 2 shows the per-source breakdown) . Sources are weighted equally, and documents equally within a source; shaded regions show 95% two-stage cluster-bootstrap CIs. Both countries sit near zero in 2021 and rise from 2024 on.  \ntions would bear on a wide range of proposed governance interventions. AI use may also be informative about attitudes toward AI itself: experimental work on civilians shows that exposure to LLMs causally shifts policy preferences (Haslberger et al., 2025), and survey work finds (intuitively) that using AI predicts attitudes toward its use (Chen & Jia, 2026) .1  \nThe instruments we currently rely on for these questions are valuable but partial. Public strategies and policy statements describe official priorities, but can lag practice and emphasize politically salient uses (U.S. Government Accountability Office, 2025 ; Blomquist, 2026) . Procurement records, staffing patterns, budget lines, and agency announcements add detail, but are noisy and difficult to compare across countries (Johnson et al., 2025 ; Haag, 2025) .  \nWe therefore propose an additional signal: traces of language-model assistance in the documents that governments publish in the course of their work. This signal is cheap to collect, externally reproducible, and—importantly—reflects revealed behavior rather than stated intent. We call it a monit","cbCailvUEDivWUHU","https://ap.wps.com/l/cbCailvUEDivWUHU","pdf",1432261,1,34,"English","en",105,"# Abstract\n# One Motivation: Indirect Traces in Public Data\n# Existing Instruments and Their Limits\n# Proposed Monitoring Primitive\n# Our Contributions\n# Data and Method\n## Data collection and sources\n## Baseline and comparison windows\n## AI-text detection with Pangram\n# Results and Signal Patterns\n# Discussion: Fit, Limitations, and Open Problems","[{\"question\":\"Why is measuring government AI use difficult with current approaches?\",\"answer\":\"Procurement disclosures and official statements can be delayed or selective and often better capture formal adoption than real day-to-day writing practices. Other instruments are detailed yet noisy and hard to compare across countries.\"},{\"question\":\"What monitoring primitive does the document propose?\",\"answer\":\"It proposes measuring traces of language-model assistance in public government documents. The signal is intended to reflect revealed behavior using a consistent, lightweight computation.\"},{\"question\":\"What does the pilot study find about AI-assisted writing signals over time?\",\"answer\":\"Signals are consistently near zero around 2021, then rise by 2026. Four of ten sources show statistically significant signs of AI-assisted writing, with the concentration pattern differing between U.S. and PRC-related streams.\"}]",1784190652,86,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":85,"head_meta":87,"extra_data":89,"updated_unix":27},"government-ai-use-as-a-monitoring-primitive-a-public-document-pilot-study","",{"@graph":35,"@context":84},[36,53,67],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/government-ai-use-as-a-monitoring-primitive-a-public-document-pilot-study/83823/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":61,"encodingFormat":60,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":4},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"Why is measuring government AI use difficult with current approaches?","Question",{"text":74,"@type":75},"Procurement disclosures and official statements can be delayed or selective and often better capture formal adoption than real day-to-day writing practices. Other instruments are detailed yet noisy and hard to compare across countries.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"What monitoring primitive does the document propose?",{"text":79,"@type":75},"It proposes measuring traces of language-model assistance in public government documents. The signal is intended to reflect revealed behavior using a consistent, lightweight computation.",{"name":81,"@type":72,"acceptedAnswer":82},"What does the pilot study find about AI-assisted writing signals over time?",{"text":83,"@type":75},"Signals are consistently near zero around 2021, then rise by 2026. Four of ten sources show statistically significant signs of AI-assisted writing, with the concentration pattern differing between U.S. and PRC-related streams.","https://schema.org",{"og:url":51,"og:type":86,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":88,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,119,122,127,130,134],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":45,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":105,"slug":137},19,"General","general"]