pith. sign in

arxiv: 2606.28331 · v1 · pith:ZTYHSFIKnew · submitted 2026-05-25 · 💻 cs.CY · cs.AI

"AI Watermarking": Bridging Policy Discourse and Technical Capabilities

Andr\'es F\'abrega , Arkaprabha Bhattacharya , Miranda Christ , Sunoo Park This is my paper

Pith reviewed 2026-06-30 11:31 UTC · model grok-4.3

classification 💻 cs.CY cs.AI
keywords AI watermarkingpolicy analysisgenerative AIcontent detectionUS legislationEU policyregulatory gapsinductive coding
0
0 comments X

The pith

Policy proposals for tracking AI-generated content assume detection capabilities that existing methods cannot reliably deliver.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The authors compile a corpus of US and EU legislative and policy documents addressing transparency for generative AI content. They apply inductive coding to identify patterns and systematically surface gaps between policy expectations and technical practice. A reader would care because these gaps could result in regulations that fail to achieve their goals or create new problems in enforcement and innovation. The analysis points to specific disconnects and ambiguities in how watermarking and similar techniques are discussed.

Core claim

The paper establishes that there are critical disconnects between the demands in policy discourse for reliable tracking of AI-generated content and the actual capabilities and limitations of current technical methods like watermarking, while also surfacing ambiguities and potential pitfalls in the policy language and trends.

What carries the argument

Inductive coding applied to a broad corpus of policy documents on AI content transparency to reveal patterns of disconnect.

If this is right

  • Policies may require watermarking features that are not yet technically feasible at scale.
  • Ambiguities in policy documents could lead to varied interpretations and inconsistent application across jurisdictions.
  • Trends in the discourse highlight pitfalls such as over-reliance on unproven detection methods.
  • Open questions remain about how to align future technical development with regulatory needs.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Closer integration between policy drafting and technical experts could reduce these mismatches in future regulations.
  • The identified gaps suggest that enforcement of such policies might require new standards for what counts as reliable detection.
  • Similar disconnects may exist in other areas of AI governance beyond content tracking.

Load-bearing premise

The methodology for selecting documents and performing inductive coding produces a representative corpus that accurately identifies the main policy-relevant gaps without bias.

What would settle it

Finding that major policy proposals on AI content tracking are fully consistent with demonstrated technical capabilities of watermarking and detection systems would challenge the central claim of significant disconnects.

Figures

Figures reproduced from arXiv: 2606.28331 by Andr\'es F\'abrega, Arkaprabha Bhattacharya, Miranda Christ, Sunoo Park.

Figure 1
Figure 1. Figure 1: Relationship between example terms that appear in [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
read the original abstract

The widespread deployment of generative artificial intelligence (AI) models has raised serious concerns about the proliferation of AI-generated content. This has led to a surge of interest in, and demand for, reliable tracking and detection mechanisms for content that is AI-generated, such as watermarking, metadata tagging, content tagging, and more. The problem has captured the attention of policymakers as well as the popular media, and a spate of recent bills in the US have sought to regulate the spread of AI content, and enforce or promote methods to track and label it. This work performs a critical analysis of the policy discourse surrounding generative AI content transparency in the US and EU. Through a broad document selection methodology, we first collect a broad corpus of documents containing legislative language and policy-relevant discourse on the topic. We then analyze these through inductive coding, and leverage our coding to systematize these documents, identifying key patterns, gaps, and open questions. We identify critical points of disconnect between policy and technological capabilities and practice, and we highlight and discuss potential ambiguities and pitfalls raised by the trends in our corpus.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The paper performs a critical analysis of policy discourse on generative AI content transparency in the US and EU. It collects a broad corpus of legislative and policy-relevant documents via a broad document selection methodology, applies inductive coding to systematize the documents and surface patterns/gaps/open questions, and identifies critical disconnects between policy language and technical capabilities along with ambiguities and pitfalls in the trends observed.

Significance. If the corpus is representative and the coding reliable, the work could usefully map mismatches between policy ambitions for AI content tracking (e.g., watermarking mandates) and current technical practice, providing a foundation for more grounded regulation. The inductive approach is appropriate for surfacing emergent themes in policy documents.

major comments (1)
  1. [Methodology] Methodology (document collection and inductive coding): the description supplies no corpus size, explicit search strings, inclusion/exclusion criteria, or inter-coder reliability statistics. Because the central claims rest on the coded corpus reliably surfacing representative disconnects and pitfalls, the absence of these details prevents assessment of selection or interpretation bias.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive review and for recognizing the potential value of mapping policy-technical disconnects in AI content transparency. We agree that greater methodological transparency is needed to support assessment of the corpus and coding process, and we will revise the manuscript to address this.

read point-by-point responses

  1. Referee: [Methodology] Methodology (document collection and inductive coding): the description supplies no corpus size, explicit search strings, inclusion/exclusion criteria, or inter-coder reliability statistics. Because the central claims rest on the coded corpus reliably surfacing representative disconnects and pitfalls, the absence of these details prevents assessment of selection or interpretation bias.

    Authors: We agree this information is necessary for evaluating selection and interpretation bias. In the revised manuscript we will expand the methodology section to report the exact corpus size, the explicit search strings and sources used for document collection, the full inclusion/exclusion criteria, and a description of the inductive coding process including how consistency was maintained (e.g., team review and consensus procedures). These additions will directly address the concern while preserving the qualitative, inductive nature of the study. revision: yes

Circularity Check

0 steps flagged

No circularity: qualitative policy analysis with no derivations or self-referential constructions

full rationale

The paper conducts a critical analysis of policy documents via broad corpus collection and inductive coding. It contains no equations, fitted parameters, predictions, uniqueness theorems, or ansatzes. All enumerated circularity patterns require self-definition, fitted inputs called predictions, or load-bearing self-citations that reduce a central claim to its own inputs; none are present. The methodology describes document selection and coding at a high level but does not invoke any internal construction that forces the reported disconnects. The analysis is therefore self-contained against external policy texts.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a qualitative policy-analysis paper containing no mathematical derivations, fitted parameters, or postulated entities.

pith-pipeline@v0.9.1-grok · 5728 in / 1022 out tokens · 32778 ms · 2026-06-30T11:31:46.099884+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

74 extracted references · 4 canonical work pages · 1 internal anchor

  1. [1]

    https://gith ub.com/hackerfactor/SEAL .PERMALINK: https: //perma.cc/BPZ5-CKUE

    Secure evidence attribution label (seal). https://gith ub.com/hackerfactor/SEAL .PERMALINK: https: //perma.cc/BPZ5-CKUE

  2. [2]

    My AI Safety Lecture for UT Effective Altruism

    Scott Aaronson. My AI Safety Lecture for UT Effective Altruism. https://scottaaronson.blog/?p=6823 , 2022.PERMALINK:https://perma.cc/F8P2-BRPP

    2022

  3. [3]

    AI could empower and prolif- erate social engineering cyberattacks

    Abdulaziz Almaslukh. AI could empower and prolif- erate social engineering cyberattacks. https://ww w.weforum.org/stories/2024/10/ai- agent s-in-cybersecurity-the-augmented-risks-w e-all-need-to-know-about/ , 2024.PERMALINK: https://perma.cc/3NFC-8FNN

    2024

  4. [4]

    Regulatory Mapping on Artificial Intel- ligence in Latin America

    Access Now. Regulatory Mapping on Artificial Intel- ligence in Latin America. https://www.accessno w.org/wp-content/uploads/2024/07/TRF-LAC -Reporte-Regional-IA-JUN-2024-V3.pdf , 2024. PERMALINK:https://perma.cc/2LHM-FG83

    2024

  5. [5]

    The simple macroeconomics of AI

    Daron Acemoglu. The simple macroeconomics of AI. Economic Policy, 2025

    2025

  6. [6]

    Do the Right Thing

    Kendra Albert and James Grimmelmann. Do the Right Thing. Communications of the ACM, 2023

    2023

  7. [7]

    The Real Dangers of Generative AI.Journal of Democracy, 2024

    Danielle Allen and E Glen Weyl. The Real Dangers of Generative AI.Journal of Democracy, 2024

    2024

  8. [8]

    Amazon Titan Image Generator and watermark detection API are now available in Ama- zon Bedrock

    Antje Barth. Amazon Titan Image Generator and watermark detection API are now available in Ama- zon Bedrock. https://aws.amazon.com/blo gs/aws/amazon- titan- image- generator- and - watermark- detection- api- are- now- avail able- in- amazon- bedrock/ , 2025.PERMALINK: https://perma.cc/THH8-C9AG

    2025

  9. [9]

    Epistemological per- spectives on is research: a framework for analysing and systematizing epistemological assumptions.Informa- tion Systems Journal, 17(2):197–214, 2007

    Jörg Becker and Björn Niehaves. Epistemological per- spectives on is research: a framework for analysing and systematizing epistemological assumptions.Informa- tion Systems Journal, 17(2):197–214, 2007

    2007

  10. [10]

    low rate of accuracy

    Benj Edwards. OpenAI discontinues its AI writing detector due to “low rate of accuracy”’. https://ar stechnica.com/information-technology/2023 /07/openai-discontinues-its-ai-writing-d etector-due-to-low-rate-of-accuracy/ , 2023. PERMALINK:https://perma.cc/4PD9-WC92

    2023

  11. [11]

    Coalition for Content provenance and authen- ticity

    C2PA. Coalition for Content provenance and authen- ticity. https://c2pa.org/ .PERMALINK: https: //perma.cc/3J4Q-HKQT

  12. [12]

    Anna Carobene, Andrea Padoan, Federico Cabitza, Giuseppe Banfi, and Mario Plebani. Rising adoption of artificial intelligence in scientific publishing: evaluating the role, risks, and ethical implications in paper drafting and review process.Clinical Chemistry and Laboratory Medicine (CCLM), 2024

    2024

  13. [13]

    Generative AI and Electoral Commu- nications.Georgetown Law Technology Review, 2025

    Evan Chiacchiaro. Generative AI and Electoral Commu- nications.Georgetown Law Technology Review, 2025

    2025

  14. [14]

    Nist’s software un-standards.George- town Law Tech Review, 2025

    Bryan H Choi. Nist’s software un-standards.George- town Law Tech Review, 2025

    2025

  15. [15]

    Pseudorandom Error- Correcting Codes

    Miranda Christ and Sam Gunn. Pseudorandom Error- Correcting Codes. InCRYPTO, 2024

    2024

  16. [16]

    Undetectable Watermarks for Language Models

    Miranda Christ, Sam Gunn, and Or Zamir. Undetectable Watermarks for Language Models. InPMLR, 2024

    2024

  17. [17]

    Securing the future of GenAI: Policy and technology.arXiv preprint arXiv:2407.12999, 2024

    Mihai Christodorescu, Ryan Craven, Soheil Feizi, Neil Gong, Mia Hoffmann, Somesh Jha, Zhengyuan Jiang, Mehrdad Saberi Kamarposhti, John Mitchell, Jessica Newman, et al. Securing the future of GenAI: Policy and technology.arXiv preprint arXiv:2407.12999, 2024

    work page arXiv 2024

  18. [18]

    Watermarking Language Models for Many Adaptive Users

    Aloni Cohen, Alexander Hoover, and Gabe Schoenbach. Watermarking Language Models for Many Adaptive Users. InIEEE S&P, 2025

    2025

  19. [19]

    How ChatGPT Could Embed a ‘Water- mark’ in the Text It Generates.https://www.nytime s.com/interactive/2023/02/17/business/ai-t ext-detection.html, 2023

    Keith Collins. How ChatGPT Could Embed a ‘Water- mark’ in the Text It Generates.https://www.nytime s.com/interactive/2023/02/17/business/ai-t ext-detection.html, 2023. 22

    2023

  20. [20]

    Provisions on the Administration of Deep Synthesis Internet Information Services

    Cyberspace Administration of China. Provisions on the Administration of Deep Synthesis Internet Information Services. https://www.chinalawtranslate.com/ en/deep-synthesis/ , 2022.PERMALINK: https: //perma.cc/XGH9-XHWD

    2022

  21. [21]

    Tay- lan Cemgil, Zahra Ahmed, Kitty Stacpoole, Ilia Shu- mailov, Ciprian Baetu, Sven Gowal, Demis Hassabis, and Pushmeet Kohli

    Sumanth Dathathri, Abigail See, Sumedh Ghaisas, Po- Sen Huang, Rob McAdam, Johannes Welbl, Vandana Bachani, Alex Kaskasoli, Robert Stanforth, Tatiana Matejovicova, Jamie Hayes, Nidhi Vyas, Majd Al Merey, Jonah Brown-Cohen, Rudy Bunel, Borja Balle, A. Tay- lan Cemgil, Zahra Ahmed, Kitty Stacpoole, Ilia Shu- mailov, Ciprian Baetu, Sven Gowal, Demis Hassabis...

    2024

  22. [22]

    Report on encryption, anonymity, and the human rights framework

    David Kaye. Report on encryption, anonymity, and the human rights framework. https://docs.un.org/A/ HRC/29/32, 2015.PERMALINK: https://perma.cc /EP3N-JYLF

    2015

  23. [23]

    Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality

    Fabrizio Dell’Acqua, Edward McFowland III, Ethan R Mollick, Hila Lifshitz-Assaf, Katherine Kellogg, Saran Rajendran, Lisa Krayer, François Candelon, and Karim R Lakhani. Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality. Harvard Business School Technology & Operation...

    2023

  24. [24]

    Evaluating the efficacy of AI content detection tools in differentiating between human and AI-generated text

    Ahmed M Elkhatat, Khaled Elsaid, and Saeed Almeer. Evaluating the efficacy of AI content detection tools in differentiating between human and AI-generated text. International Journal for Educational Integrity, 2023

    2023

  25. [25]

    GPTs are GPTs: Labor market impact potential of LLMs.Science, 2024

    Tyna Eloundou, Sam Manning, Pamela Mishkin, and Daniel Rock. GPTs are GPTs: Labor market impact potential of LLMs.Science, 2024

    2024

  26. [26]

    Publicly-Detectable Watermarking for Language Models.IACR Communications in Cryptology, 2024

    Jaiden Fairoze, Sanjam Garg, Somesh Jha, Saeed Mahloujifar, Mohammad Mahmoody, and Mingyuan Wang. Publicly-Detectable Watermarking for Language Models.IACR Communications in Cryptology, 2024

    2024

  27. [27]

    Demon- strating Rigor Using Thematic Analysis: A Hybrid Ap- proach of Inductive and Deductive Coding and Theme Development.International Journal of Qualitative Methods, 2006

    Jennifer Fereday and Eimear Muir-Cochrane. Demon- strating Rigor Using Thematic Analysis: A Hybrid Ap- proach of Inductive and Deductive Coding and Theme Development.International Journal of Qualitative Methods, 2006

    2006

  28. [28]

    Labeling Synthetic Content: User Perceptions of Label Designs for AI-Generated Content on Social Media

    Dilrukshi Gamage, Dilki Sewwandi, Min Zhang, and Arosha K Bandara. Labeling Synthetic Content: User Perceptions of Label Designs for AI-Generated Content on Social Media. InCHI, 2025

    2025

  29. [29]

    Alexander Gamero-Garrido, Stefan Savage, Kirill Levchenko, and Alex C. Snoeren. Quantifying the Pres- sure of Legal Risks on Third-party Vulnerability Re- search. In Bhavani Thuraisingham, David Evans, Tal Malkin, and Dongyan Xu, editors,ACM CCS

  30. [30]

    Two AI Transparency Concerns that Governments Should Align On

    Tommaso Giardini, Nora Fischer, and Johannes Fritz. Two AI Transparency Concerns that Governments Should Align On. https://www.techpolicy.pre ss/two-ai-transparency-concerns-that-gov ernments-should-align-on/ , 2024.PERMALINK: https://perma.cc/9YML-PNK4

    2024

  31. [31]

    SynthID – A tool to watermark and identify content generated through AI

    Google DeepMind. SynthID – A tool to watermark and identify content generated through AI. https://de epmind.google/science/synthid/ .PERMALINK: https://perma.cc/Z5AW-XNNV

  32. [32]

    Greenberg

    Brad A. Greenberg. Rethinking Technology Neutrality. Minnesota Law Review, 2015

    2015

  33. [33]

    Reg- ulating ChatGPT and other large generative AI models

    Philipp Hacker, Andreas Engel, and Marco Mauer. Reg- ulating ChatGPT and other large generative AI models. InACM FAccT, 2023

    2023

  34. [34]

    Global Standard Setting in Internet Governance

    Alison Harcourt, George Christou, and Seamus Simpson. Global Standard Setting in Internet Governance. Oxford University Press, 2020

    2020

  35. [35]

    Center for Countering Digital Hate. The Double-Edged Sword of AI: How Generative Language Models Like Google Bard and ChatGPT Pose a Threat to Countering Hate and Misinformation Online.Harvard Data Science Review, 2024

    2024

  36. [36]

    RE: Senate Bill Number 97 of the 2024 Regular Session by Senator Royce Duplessis

    Jeff Landry. RE: Senate Bill Number 97 of the 2024 Regular Session by Senator Royce Duplessis. https: //www.legis.la.gov/legis/ViewDocument.aspx ?d=1382564, 2024.PERMALINK: https://perma.cc /WB44-9RNP

    2024

  37. [37]

    Evading Watermark based Detection of AI- Generated Content

    Zhengyuan Jiang, Jinghuai Zhang, and Neil Zhenqiang Gong. Evading Watermark based Detection of AI- Generated Content. InACM CCS, 2023

    2023

  38. [38]

    Nikola Jovanovic, Robin Staab, and Martin T. Vechev. Watermark Stealing in Large Language Models. In ICML, 2024

    2024

  39. [39]

    A Watermark for Large Language Models

    John Kirchenbauer, Jonas Geiping, Yuxin Wen, Jonathan Katz, Ian Miers, and Tom Goldstein. A Watermark for Large Language Models. In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett, editors,ICML, 2023

    2023

  40. [40]

    Robust Distortion-free Watermarks for Language Models.TMLR, 2024

    Rohith Kuditipudi, John Thickstun, Tatsunori Hashimoto, and Percy Liang. Robust Distortion-free Watermarks for Language Models.TMLR, 2024. 23

    2024

  41. [41]

    How we’re increasing transparency for gen AI content with the C2PA

    Laurie Richardson. How we’re increasing transparency for gen AI content with the C2PA. https://blog .google/technology/ai/google- gen- ai- con tent- transparency- c2pa/ , 2024.PERMALINK: https://perma.cc/3S98-9CBW

    2024

  42. [42]

    An empirical investigation of the impact of ChatGPT on creativity

    Byung Cheol Lee and Jaeyeon Chung. An empirical investigation of the impact of ChatGPT on creativity. Nature Human Behaviour, 2024

    2024

  43. [43]

    GPT detectors are biased against non- native English writers.Patterns, 2023

    Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, and James Zou. GPT detectors are biased against non- native English writers.Patterns, 2023

    2023

  44. [44]

    ChatGPT Hits 1 Billion Users? ‘Doubled In Just Weeks’ Says OpenAI CEO

    Martine Paris. ChatGPT Hits 1 Billion Users? ‘Doubled In Just Weeks’ Says OpenAI CEO. https://www.fo rbes.com/sites/martineparis/2025/04/12/cha tgpt-hits-1-billion-users-openai-ceo-say s-doubled-in-weeks/ , 2025.PERMALINK: https: //perma.cc/2UF2-7ATN

    2025

  45. [45]

    One of the Biggest Problems in Regulating AI Is Agreeing on a Definition

    Matt O’Shaughnessy. One of the Biggest Problems in Regulating AI Is Agreeing on a Definition. https: //carnegieendowment.org/posts/2022/10/one-o f-the-biggest-problems-in-regulating-ai-i s-agreeing-on-a-definition , 2022.PERMALINK: https://perma.cc/5RUV-RWRC

    2022

  46. [46]

    Google’s Gemini AI app has 400M monthly active users

    Maxwell Zeff. Google’s Gemini AI app has 400M monthly active users. https://techcrunch.com /2025/05/20/googles- gemini- ai- app- has- 4 00m-monthly-active-users , 2025.PERMALINK: https://perma.cc/P5M3-D8V9

    2025

  47. [47]

    Interrater reliability: the kappa statis- tic.Biochemia medica, 2012

    Mary L McHugh. Interrater reliability: the kappa statis- tic.Biochemia medica, 2012

    2012

  48. [48]

    Detect- GPT: Zero-Shot Machine-Generated Text Detection us- ing Probability Curvature

    Eric Mitchell, Yoonho Lee, Alexander Khazatsky, Christopher D Manning, and Chelsea Finn. Detect- GPT: Zero-Shot Machine-Generated Text Detection us- ing Probability Curvature. InICML, 2023

    2023

  49. [49]

    Content Industry Promotion Act, 2024

    National Assembly of South Korea. Content Industry Promotion Act, 2024

    2024

  50. [50]

    Watermarking Without Standards Is Not AI Governance

    Alexander Nemecek, Yuzhou Jiang, and Erman Ayday. Watermarking Without Standards Is Not AI Governance. arXiv preprint arXiv:2505.23814, 2025

    work page arXiv 2025

  51. [51]

    Labeling AI-Generated Images on Face- book, Instagram and Threads

    Nick Clegg. Labeling AI-Generated Images on Face- book, Instagram and Threads. https://about.fb .com/news/2024/02/labeling-ai-generated-i mages-on-facebook-instagram-and-threads/ , 2024.PERMALINK:https://perma.cc/2PHP-GKXT

    2024

  52. [52]

    Introducing ChatGPT

    OpenAI. Introducing ChatGPT. https://openai .com/index/chatgpt/ , 2022.PERMALINK: https: //perma.cc/S23B-THHG

    2022

  53. [53]

    Understanding the source of what we see and hear online

    OpenAI. Understanding the source of what we see and hear online. https://openai.com/index/underst anding-the-source-of-what-we-see-and-hea r-online/, 2024.PERMALINK: https://perma.cc /W55G-CV8G

    2024

  54. [54]

    A Researcher’s Guide to Some Legal Risks of Security Research

    Sunoo Park and Kendra Albert. A Researcher’s Guide to Some Legal Risks of Security Research. A joint pub- lication of the Cyberlaw Clinic at Harvard Law School, the Technology Law and Policy Clinic at NYU Law, and the Electronic Frontier Foundation, 2024. Available at: https://clinic.cyber.harvard.edu/wp-conte nt/uploads/2024/08/Security-Researchers-G uid...

    2024

  55. [55]

    Misunderstanding AI’s Democracy Problem.UCLA Journal of Law & Technology, 2025

    Nathaniel Persily. Misunderstanding AI’s Democracy Problem.UCLA Journal of Law & Technology, 2025

    2025

  56. [56]

    Taking Sides on Technology Neutrality

    Chris Reed. Taking Sides on Technology Neutrality. SCRIPTed, 2007

    2007

  57. [57]

    Adoption of Watermarking Measures for AI-Generated Content and Implications under the EU AI Act.arXiv preprint arXiv:2503.18156, 2025

    Bram Rijsbosch, Gijs van Dijck, and Konrad Kollnig. Adoption of Watermarking Measures for AI-Generated Content and Implications under the EU AI Act.arXiv preprint arXiv:2503.18156, 2025

    work page arXiv 2025

  58. [58]

    SAGE publications, 2014

    Johnny Saldana.Thinking Qualitatively: Methods of Mind. SAGE publications, 2014

    2014

  59. [59]

    OpenAI has the tech to watermark ChatGPT text–it just won’t release it

    Samuel Axon. OpenAI has the tech to watermark ChatGPT text–it just won’t release it. https://ar stechnica.com/ai/2024/08/openai- has- the - tech- to- watermark- chatgpt- text- it- jus t-wont-release-it/ , 2024.PERMALINK: https: //perma.cc/4F54-HS5G

    2024

  60. [60]

    Watermarking: where to? https: //www.youtube.com/live/YzuVet3YkkA?si=p0QG bUn07Rx-zO-W, 2024.PERMALINK: https://perma

    Scott Aaronson. Watermarking: where to? https: //www.youtube.com/live/YzuVet3YkkA?si=p0QG bUn07Rx-zO-W, 2024.PERMALINK: https://perma. cc/2SUK-DHGT

    2024

  61. [61]

    There’s a Tool to Catch Students Cheating With ChatGPT

    Deepa Seetharaman and Matt Barnum. There’s a Tool to Catch Students Cheating With ChatGPT. OpenAI Hasn’t Released It. https://www.wsj.com/tech/ai/open ai-tool-chatgpt-cheating-writing-135b755a , 2024

    2024

  62. [62]

    AI models collapse when trained on recursively generated data

    Ilia Shumailov, Zakhar Shumaylov, Yiren Zhao, Nicolas Papernot, Ross Anderson, and Yarin Gal. AI models collapse when trained on recursively generated data. Nature, 2024

    2024

  63. [63]

    AI on Snapchat: Improved Transparency, Safety, and Policies

    Snap Values. AI on Snapchat: Improved Transparency, Safety, and Policies. https://values.snap.com/ne ws/ai-on-snapchat-improved-transparency-s afety-policy, 2024.PERMALINK: https://perma. cc/XB7W-X3ZH. 24

    2024

  64. [64]

    A Real Account of Deep Fakes.Cor- nell Legal Studies Research Paper Forthcoming, Michi- gan Law Review, Forthcoming, 2024

    Benjamin Sobel. A Real Account of Deep Fakes.Cor- nell Legal Studies Research Paper Forthcoming, Michi- gan Law Review, Forthcoming, 2024

    2024

  65. [65]

    Initial Rescissions of Harmful Ex- ecutive Orders and Actions

    The White House. Initial Rescissions of Harmful Ex- ecutive Orders and Actions. https://www.whit ehouse.gov/presidential- actions/2025/01 /initial- rescissions- of- harmful- executi ve- orders- and- actions/ , 2025.PERMALINK: https://perma.cc/FCN3-F6DQ

    2025

  66. [66]

    Removing Barriers to American Leadership in Artificial Intelligence

    The White House. Removing Barriers to American Leadership in Artificial Intelligence. https://www. whitehouse.gov/presidential-actions/2025/ 01/removing-barriers-to-american-leadershi p-in-artificial-intelligence/ , 2025.PERMA- LINK:https://perma.cc/324U-9K2W

    2025

  67. [67]

    LLaMA: Open and Efficient Foundation Language Models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Bap- tiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. LLaMA: Open and Efficient Foundation Language Models.arXiv preprint arXiv:2302.13971, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  68. [68]

    Dates of Sessions of the Congress

    United States Senate. Dates of Sessions of the Congress. https://www.senate.gov/legislative/Dates ofSessionsofCongress.htm .PERMALINK: https: //perma.cc/R3AJ-9JN5

  69. [69]

    Even in Death, Internet Explorer Lives On in South Korea

    Daisuke Wakabayashi and Jin Yu Young. Even in Death, Internet Explorer Lives On in South Korea. https: //www.nytimes.com/2022/07/08/business/kore a-internet-explorer.html, 2022

    2022

  70. [70]

    Internet Governance, Standard Setting, and Self-Regulation.Northern Kentucky Law Review, 2001

    Philip J Weiser. Internet Governance, Standard Setting, and Self-Regulation.Northern Kentucky Law Review, 2001

    2001

  71. [71]

    Watermarks in the Sand: Impossibility of Strong Water- marking for Large Language Models

    Hanlin Zhang, Benjamin L Edelman, Danilo Francati, Daniele Venturi, Giuseppe Ateniese, and Boaz Barak. Watermarks in the Sand: Impossibility of Strong Water- marking for Large Language Models. InICML, 2024

    2024

  72. [72]

    REMARK-LLM: A Robust and Efficient Watermarking Framework for Gen- erative Large Language Models

    Ruisi Zhang, Shehzeen Samarah Hussain, Paarth Neekhara, and Farinaz Koushanfar. REMARK-LLM: A Robust and Efficient Watermarking Framework for Gen- erative Large Language Models. InUSENIX Security, 2024

    2024

  73. [73]

    SoK: Watermarking for AI-Generated Content

    Xuandong Zhao, Sam Gunn, Miranda Christ, Jaiden Fairoze, Andres Fabrega, Nicholas Carlini, Sanjam Garg, Sanghyun Hong, Milad Nasr, Florian Tramèr, Somesh Jha, Lei Li, Yu-Xiang Wang, and Dawn Song. SoK: Watermarking for AI-Generated Content. InIEEE S&P, 2025

    2025

  74. [74]

    substantially modified

    Xuandong Zhao, Kexun Zhang, Zihao Su, Saastha Vasan, Ilya Grishchenko, Christopher Kruegel, Giovanni Vigna, Yu-Xiang Wang, and Lei Li. Invisible Image Water- marks Are Provably Removable Using Generative AI. InNeurIPS, 2024. 25 A Recap of Key Takeaways and Open Ques- tions In this section, we list all key trends and open questions iden- tified in Sections...

    2024