pith. sign in

arxiv: 2604.12216 · v1 · submitted 2026-04-14 · 💻 cs.CR · cs.CL

TimeMark: A Trustworthy Time Watermarking Framework for Exact Generation-Time Recovery from AIGC

Shangkun Che , Silin Du , Ge Gao This is my paper

Pith reviewed 2026-05-10 16:09 UTC · model grok-4.3

classification 💻 cs.CR cs.CL
keywords trustworthy watermarkingtime watermarkAIGCLLM watermarkgeneration time recoverycryptographic watermarkjudicial evidenceintellectual property
0
0 comments X

The pith

TimeMark recovers the exact generation time of AI-generated text with perfect accuracy by binding timestamps to regulated secret keys and using two-stage encoding plus error correction.

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

The paper introduces a trustworthy time watermarking framework that embeds generation timestamps into LLM outputs so they can serve as reliable judicial evidence in intellectual property disputes. Unlike prior statistical watermarking methods that only give probabilistic detection and allow forgery, this approach decouples the payload from time, generates random non-stored bits per instance, and relies on time-dependent keys supervised by regulators. A two-stage encoding step combined with error-correcting codes is claimed to deliver theoretically perfect recovery while resisting both user-side statistical attacks and provider-side fabrication. The authors argue that these properties meet the reliability bar needed for court use and provide a concrete path to resolve future AIGC disputes.

Core claim

The framework integrates cryptographic techniques to encode time information into time-dependent secret keys under regulatory supervision, preventing arbitrary timestamp fabrication. The watermark payload is generated as a random, non-stored bit sequence for each instance, eliminating statistical patterns. A two-stage encoding mechanism together with error-correcting codes enables reliable recovery of the generation time with theoretically perfect accuracy.

What carries the argument

Two-stage encoding mechanism combined with error-correcting codes that operates on random payloads derived from time-dependent secret keys under regulatory supervision.

If this is right

  • AI-generated text can be timestamped in a way that satisfies judicial standards for evidence in copyright and IP cases.
  • Model providers lose the ability to fabricate arbitrary generation times because keys are externally supervised.
  • Statistical detection attacks become ineffective because payloads carry no distributional patterns.
  • The same reliability guarantees extend to multi-bit information beyond timestamps when the two-stage encoding is applied.

Where Pith is reading between the lines

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

  • The method could be extended to image, audio, or video AIGC if analogous key-supervision and encoding steps are defined for those modalities.
  • Widespread adoption would shift the burden of timestamp integrity from technical detection to regulatory key management.
  • If perfect recovery holds, courts could treat recovered timestamps as stronger evidence than current probabilistic watermarks.
  • The framework creates an incentive for standardized regulatory infrastructure around time-key issuance.

Load-bearing premise

That regulators can reliably enforce time-dependent secret keys so providers cannot forge timestamps, and that the two-stage encoding with error correction will always recover the time perfectly from any real-world LLM output without statistical leaks or implementation errors.

What would settle it

A demonstration that a model provider can produce text containing an arbitrary future or past timestamp that still passes the recovery procedure, or a recovery failure on a set of LLM outputs where the embedded time cannot be reconstructed exactly.

read the original abstract

The widespread use of Large Language Models (LLMs) in text generation has raised increasing concerns about intellectual property disputes. Watermarking techniques, which embed meta information into AI-generated content (AIGC), have the potential to serve as judicial evidence. However, existing methods rely on statistical signals in token distributions, leading to inherently probabilistic detection and reduced reliability, especially in multi-bit encoding (e.g., timestamps). Moreover, such methods introduce detectable statistical patterns, making them vulnerable to forgery attacks and enabling model providers to fabricate arbitrary watermarks. To address these issues, we propose the concept of trustworthy watermark, which achieves reliable recovery with 100% identification accuracy while resisting both user-side statistical attacks and provider-side forgery. We focus on trustworthy time watermarking for use as judicial evidence. Our framework integrates cryptographic techniques and encodes time information into time-dependent secret keys under regulatory supervision, preventing arbitrary timestamp fabrication. The watermark payload is decoupled from time and generated as a random, non-stored bit sequence for each instance, eliminating statistical patterns. To ensure verifiability, we design a two-stage encoding mechanism, which, combined with error-correcting codes, enables reliable recovery of generation time with theoretically perfect accuracy. Both theoretical analysis and experiments demonstrate that our framework satisfies the reliability requirements for judicial evidence and offers a practical solution for future AIGC-related intellectual property disputes.

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

2 major / 0 minor

Summary. The paper proposes TimeMark, a trustworthy time watermarking framework for exact generation-time recovery from AI-generated content (AIGC). It integrates cryptographic techniques with time-dependent secret keys under regulatory supervision to prevent forgery, uses a decoupled random payload to avoid statistical patterns, and employs a two-stage encoding mechanism combined with error-correcting codes to achieve theoretically perfect accuracy in time recovery. The authors claim that theoretical analysis and experiments show it meets reliability requirements for judicial evidence in AIGC intellectual property disputes, resisting both user-side attacks and provider-side forgery.

Significance. If the results hold, the framework could offer a substantial improvement over existing probabilistic watermarking methods by providing deterministic and highly reliable time stamping for AI content, which is critical for legal and IP applications. The cryptographic approach and attack resistance could set a new standard for trustworthy AIGC authentication.

major comments (2)
  1. Abstract: The assertion that the two-stage encoding mechanism combined with error-correcting codes 'enables reliable recovery of generation time with theoretically perfect accuracy' is made without any supporting equations, formal proofs, or error analysis. This is load-bearing for the central claim of 100% accuracy and judicial suitability, as no derivation is provided to show invariance to model-specific statistical properties or implementation details.
  2. Abstract: The framework's resistance to provider-side forgery is attributed to encoding time into 'time-dependent secret keys under regulatory supervision,' but this relies on an external, non-technical assumption of effective regulatory enforcement across providers. Without technical mechanisms to enforce or verify key usage, this does not constitute a cryptographic guarantee and weakens the 'trustworthy' and forgery-resistance claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the two major comments point by point below, providing clarifications on the supporting analysis and the assumptions in our framework. Where appropriate, we indicate revisions to strengthen the presentation.

read point-by-point responses

  1. Referee: Abstract: The assertion that the two-stage encoding mechanism combined with error-correcting codes 'enables reliable recovery of generation time with theoretically perfect accuracy' is made without any supporting equations, formal proofs, or error analysis. This is load-bearing for the central claim of 100% accuracy and judicial suitability, as no derivation is provided to show invariance to model-specific statistical properties or implementation details.

    Authors: The abstract provides a concise summary of the central claim. The full manuscript contains the supporting theoretical analysis, including the formal description of the two-stage encoding process, the integration with error-correcting codes, the derivation showing deterministic recovery independent of the underlying model's token statistics, and the error probability bounds (see Sections 3.2 and 4). These establish invariance to model-specific properties under the stated cryptographic assumptions and yield theoretically perfect accuracy (zero decoding error with overwhelming probability for the chosen parameters). To improve clarity, we will revise the abstract to include a brief pointer to the theoretical guarantees in the main text. revision: partial

  2. Referee: Abstract: The framework's resistance to provider-side forgery is attributed to encoding time into 'time-dependent secret keys under regulatory supervision,' but this relies on an external, non-technical assumption of effective regulatory enforcement across providers. Without technical mechanisms to enforce or verify key usage, this does not constitute a cryptographic guarantee and weakens the 'trustworthy' and forgery-resistance claims.

    Authors: We agree that the forgery resistance is not a purely cryptographic guarantee in isolation. The technical component ensures that, given a time-dependent key, forging a valid watermark for an arbitrary timestamp is computationally infeasible due to the cryptographic binding and the non-stored random payload. The regulatory supervision is an explicit assumption of the threat model (as stated in Section 2), analogous to the trusted setup in many cryptographic protocols. We do not claim a self-enforcing technical mechanism that replaces regulation. We will add a dedicated paragraph in the discussion section clarifying this assumption, its scope, and the resulting security guarantees conditional on proper key management. revision: partial

Circularity Check

0 steps flagged

No circularity detected; claims rest on external assumptions and standard primitives

full rationale

The abstract and description present the framework as integrating cryptographic techniques, time-dependent secret keys under regulatory supervision, decoupled random payloads, two-stage encoding, and error-correcting codes to achieve 100% recovery accuracy. No equations, definitions, or self-citations are provided that reduce the accuracy claim or trustworthiness to a fitted parameter, self-referential quantity, or prior author result by construction. The central claims do not exhibit self-definition, fitted inputs renamed as predictions, or load-bearing self-citation chains; they build on independent cryptographic and coding primitives with external non-technical assumptions.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

Review based on abstract only; full details on any fitted parameters or additional assumptions are unavailable.

axioms (2)
  • domain assumption Cryptographic techniques under regulatory supervision can encode time information into time-dependent secret keys that prevent arbitrary fabrication.
    Invoked to resist provider-side forgery attacks.
  • ad hoc to paper The two-stage encoding mechanism combined with error-correcting codes enables reliable recovery of generation time with theoretically perfect accuracy.
    Central assumption supporting the 100% identification accuracy claim.
invented entities (1)
  • Trustworthy time watermark no independent evidence
    purpose: To achieve 100% accurate time recovery while resisting statistical attacks and forgery.
    New concept introduced to differentiate from probabilistic statistical watermarks.

pith-pipeline@v0.9.0 · 5547 in / 1381 out tokens · 42100 ms · 2026-05-10T16:09:32.286605+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

51 extracted references · 51 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter doi edition editor eid howpublished institution isbn issn journal key month note number organization pages publisher school series title type url volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1...

    work page

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in "" FUNCTION format.date year ...

    work page

  3. [3]

    RFC 3161, ://www.ietf.org/rfc/rfc3161.txt

    Adams C, Cain P, Pinkas D, Zuccherato R (2001) Internet x.509 public key infrastructure time-stamp protocol (tsp). RFC 3161, ://www.ietf.org/rfc/rfc3161.txt

    work page 2001

  4. [4]

    Management Science

    Bick A, Blandin A, Deming DJ (2026) The rapid adoption of generative ai. Management Science

    work page 2026

  5. [5]

    Mathematics 12(23):3860

    Bulgakov AL, Aleshina AV, Smirnov SD, Demidov AD, Milyutin MA, Xin Y (2024) Scalability and security in blockchain networks: Evaluation of sharding algorithms and prospects for decentralized data storage. Mathematics 12(23):3860

    work page 2024

  6. [6]

    arXiv preprint arXiv:2409.02968

    Chen B, Ma L, Xu H, Ma J, Hu D, Liu X, Wu J, Wang J, Li K (2024) A comprehensive survey of blockchain scalability: Shaping inner-chain and inter-chain perspectives. arXiv preprint arXiv:2409.02968

    work page arXiv 2024

  7. [7]

    The Thirty Seventh Annual Conference on Learning Theory, 1125--1139 (PMLR)

    Christ M, Gunn S, Zamir O (2024) Undetectable watermarks for language models. The Thirty Seventh Annual Conference on Learning Theory, 1125--1139 (PMLR)

    work page 2024

  8. [8]

    https://c2pa.org/specifications/specifications/2.4/specs/C2PA_Specification.html, official technical specification for Content Credentials

    Coalition for Content Provenance and Authenticity (2025) C2pa technical specification, version 2.4. https://c2pa.org/specifications/specifications/2.4/specs/C2PA_Specification.html, official technical specification for Content Credentials

    work page 2025

  9. [9]

    https://rm.coe.int/guidelines-on-electronic-evidence-and-explanatory-memorandum/1680968ab5, adopted by the Committee of Ministers on January 30, 2019

    Council of Europe (2019) Guidelines on electronic evidence in civil and administrative proceedings. https://rm.coe.int/guidelines-on-electronic-evidence-and-explanatory-memorandum/1680968ab5, adopted by the Committee of Ministers on January 30, 2019

    work page arXiv 2019

  10. [10]

    https://www.cac.gov.cn/2023-07/13/c_1690898327029107.htm, promulgated July 13, 2023

    Cyberspace Administration of China and Other Agencies (2023) Interim measures for the management of generative artificial intelligence services. https://www.cac.gov.cn/2023-07/13/c_1690898327029107.htm, promulgated July 13, 2023

    work page 2023

  11. [11]

    https://www.cac.gov.cn/2025-03/14/c_1743654684782215.htm, promulgated March 14, 2025

    Cyberspace Administration of China and Other Agencies (2025) Measures for the labeling of ai-generated synthetic content. https://www.cac.gov.cn/2025-03/14/c_1743654684782215.htm, promulgated March 14, 2025

    work page 2025

  12. [12]

    (2024) Scalable watermarking for identifying large language model outputs

    Dathathri S, See A, Ghaisas S, Huang PS, McAdam R, Welbl J, Bachani V, Kaskasoli A, Stanforth R, Matejovicova T, et al. (2024) Scalable watermarking for identifying large language model outputs. Nature 634(8035):818--823

    work page 2024

  13. [13]

    Optimizing adaptive attacks against content watermarks for language models.arXiv preprint arXiv:2410.02440,

    Diaa A, Aremu T, Lukas N (2024) Optimizing adaptive attacks against watermarks for language models. arXiv preprint arXiv:2410.02440

    work page arXiv 2024

  14. [14]

    https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai, published July 10, 2025

    European Commission (2025 a ) The general-purpose ai code of practice. https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai, published July 10, 2025

    work page 2025

  15. [15]

    https://digital-strategy.ec.europa.eu/en/faqs/general-purpose-ai-models-ai-act-questions-answers, updated July 10, 2025

    European Commission (2025 b ) General-purpose ai models in the ai act -- questions & answers. https://digital-strategy.ec.europa.eu/en/faqs/general-purpose-ai-models-ai-act-questions-answers, updated July 10, 2025

    work page 2025

  16. [16]

    https://eur-lex.europa.eu/eli/reg/2014/910/oj/eng, consolidated version available through EUR-Lex

    European Union (2014) Regulation (eu) no 910/2014 on electronic identification and trust services for electronic transactions in the internal market (eidas regulation). https://eur-lex.europa.eu/eli/reg/2014/910/oj/eng, consolidated version available through EUR-Lex

    work page 2014

  17. [17]

    https://eur-lex.europa.eu/eli/dir/2019/790/oj/eng, official Journal of the European Union

    European Union (2019) Directive (eu) 2019/790 of the european parliament and of the council of 17 april 2019 on copyright and related rights in the digital single market. https://eur-lex.europa.eu/eli/dir/2019/790/oj/eng, official Journal of the European Union

    work page 2019

  18. [18]

    https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng, official Journal of the European Union

    European Union (2024) Regulation (eu) 2024/1689 of the european parliament and of the council of 13 june 2024 laying down harmonised rules on artificial intelligence (artificial intelligence act). https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng, official Journal of the European Union

    work page 2024

  19. [19]

    IACR Communications in Cryptology 1(4)

    Fairoze J, Garg S, Jha S, Mahloujifar S, Mahmoody M, Wang M (2025) Publicly-detectable watermarking for language models. IACR Communications in Cryptology 1(4)

    work page 2025

  20. [20]

    Bulletin of IEEE Technical Committee on Digital Libraries (TCDL)

    Gipp B, Meuschke N, Beel J, Breitinger C (2017) Using the blockchain of cryptocurrencies for timestamping digital cultural heritage. Bulletin of IEEE Technical Committee on Digital Libraries (TCDL)

    work page 2017

  21. [21]

    Journal of Cryptology 3(2):99--111, ://link.springer.com/article/10.1007/BF00196791

    Haber S, Stornetta WS (1991) How to time-stamp a digital document. Journal of Cryptology 3(2):99--111, ://link.springer.com/article/10.1007/BF00196791

    work page doi:10.1007/bf00196791 1991

  22. [22]

    The Twelfth International Conference on Learning Representations

    Hu Z, Chen L, Wu X, Wu Y, Zhang H, Huang H (2024) Unbiased watermark for large language models. The Twelfth International Conference on Learning Representations

    work page 2024

  23. [23]

    Stealthink: A multi-bit and stealthy watermark for large language models, 2025

    Jiang Y, Wu C, Boroujeny MK, Mark B, Zeng K (2025) Stealthink: A multi-bit and stealthy watermark for large language models. arXiv preprint arXiv:2506.05502

    work page arXiv 2025

  24. [24]

    Watermark stealing in large language models

    Jovanovi \'c N, Staab R, Vechev M (2024) Watermark stealing in large language models. arXiv preprint arXiv:2402.19361

    work page arXiv 2024

  25. [25]

    International conference on machine learning, 17061--17084 (PMLR)

    Kirchenbauer J, Geiping J, Wen Y, Katz J, Miers I, Goldstein T (2023) A watermark for large language models. International conference on machine learning, 17061--17084 (PMLR)

    work page 2023

  26. [26]

    https://www.law.cornell.edu/rules/fre/rule_803, accessed April 2026

    Legal Information Institute (2026 a ) Federal rule of evidence 803: Exceptions to the rule against hearsay. https://www.law.cornell.edu/rules/fre/rule_803, accessed April 2026

    work page 2026

  27. [27]

    https://www.law.cornell.edu/rules/fre/rule_901, accessed April 2026

    Legal Information Institute (2026 b ) Federal rule of evidence 901: Authenticating or identifying evidence. https://www.law.cornell.edu/rules/fre/rule_901, accessed April 2026

    work page 2026

  28. [28]

    https://www.law.cornell.edu/rules/fre/rule_902, accessed April 2026

    Legal Information Institute (2026 c ) Federal rule of evidence 902: Evidence that is self-authenticating. https://www.law.cornell.edu/rules/fre/rule_902, accessed April 2026

    work page 2026

  29. [29]

    An unforgeable publicly verifiable watermark for large language models

    Liu A, Pan L, Hu X, Li S, Wen L, King I, Yu PS (2023) An unforgeable publicly verifiable watermark for large language models. arXiv preprint arXiv:2307.16230

    work page arXiv 2023

  30. [30]

    ACM Transactions on Storage 5(1):1--21, ://dl.acm.org/doi/10.1145/1502777.1502779

    Ma D, Tsudik G (2009) A new approach to secure logging. ACM Transactions on Storage 5(1):1--21, ://dl.acm.org/doi/10.1145/1502777.1502779

    work page doi:10.1145/1502777.1502779 2009

  31. [31]

    European Symposium on Research in Computer Security, 3--24 (Springer)

    Meng L, Chen L (2022) A blockchain-based long-term time-stamping scheme. European Symposium on Research in Computer Security, 3--24 (Springer)

    work page 2022

  32. [32]

    Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, 18100--18110

    Rastogi S, Pruthi D (2024) Revisiting the robustness of watermarking to paraphrasing attacks. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, 18100--18110

    work page 2024

  33. [33]

    Findings of the Association for Computational Linguistics: NAACL 2024, 613--625

    Ren J, Xu H, Liu Y, Cui Y, Wang S, Yin D, Tang J (2024 a ) A robust semantics-based watermark for large language model against paraphrasing. Findings of the Association for Computational Linguistics: NAACL 2024, 613--625

    work page 2024

  34. [34]

    Findings of the Association for Computational Linguistics: NAACL 2024, 613--625

    Ren J, Xu H, Liu Y, Cui Y, Wang S, Yin D, Tang J (2024 b ) A robust semantics-based watermark for large language model against paraphrasing. Findings of the Association for Computational Linguistics: NAACL 2024, 613--625

    work page 2024

  35. [35]

    Siddharth E (2025) Time-stamping in blockchain for legal evidence submission. Sci. J. Artif. Intell. Blockchain Technol 2:34--42

    work page 2025

  36. [36]

    https://www.gov.uk/government/consultations/copyright-and-artificial-intelligence/copyright-and-artificial-intelligence, consultation published December 17, 2024

    UK Government (2024) Copyright and artificial intelligence. https://www.gov.uk/government/consultations/copyright-and-artificial-intelligence/copyright-and-artificial-intelligence, consultation published December 17, 2024

    work page 2024

  37. [37]

    UK Government (2026) Report on copyright and artificial intelligence. https://www.gov.uk/government/publications/report-and-impact-assessment-on-copyright-and-artificial-intelligence/report-on-copyright-and-artificial-intelligence, published March 18, 2026

    work page 2026

  38. [38]

    https://www.legislation.gov.uk/ukpga/1988/48/section/3, accessed April 2026

    United Kingdom (1988) Copyright, designs and patents act 1988, section 3. https://www.legislation.gov.uk/ukpga/1988/48/section/3, accessed April 2026

    work page 1988

  39. [39]

    United States (1976 a ) 17 u.s.c. § 101. https://www.law.cornell.edu/uscode/text/17/101, definitions; accessed April 2026

    work page 1976

  40. [40]

    United States (1976 b ) 17 u.s.c. § 102. https://www.law.cornell.edu/uscode/text/17/102, subject matter of copyright: In general; accessed April 2026

    work page 1976

  41. [41]

    ://www.uscourts.gov/sites/default/files/2025-02/federal-rules-of-evidence-dec-1-2024_0.pdf, effective Dec

    United States Courts (2024) Federal rules of evidence. ://www.uscourts.gov/sites/default/files/2025-02/federal-rules-of-evidence-dec-1-2024_0.pdf, effective Dec. 1, 2024

    work page 2024

  42. [42]

    https://www.copyright.gov/ai/, overview page for the Copyright and Artificial Intelligence report series

    US Copyright Office (2025 a ) Copyright and artificial intelligence. https://www.copyright.gov/ai/, overview page for the Copyright and Artificial Intelligence report series

    work page 2025

  43. [43]

    https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf, january 2025

    US Copyright Office (2025 b ) Copyright and artificial intelligence, part 2: Copyrightability. https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf, january 2025

    work page 2025

  44. [44]

    https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-3-Generative-AI-Training-Report-Pre-Publication-Version.pdf, may 2025

    US Copyright Office (2025 c ) Copyright and artificial intelligence, part 3: Generative ai training (pre-publication version). https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-3-Generative-AI-Training-Report-Pre-Publication-Version.pdf, may 2025

    work page 2025

  45. [45]

    The Twelfth International Conference on Learning Representations, ://openreview.net/forum?id=JYu5Flqm9D

    Wang L, Yang W, Chen D, Zhou H, Lin Y, Meng F, Zhou J, Sun X (2024) Towards codable watermarking for injecting multi-bits information to LLM s. The Twelfth International Conference on Learning Representations, ://openreview.net/forum?id=JYu5Flqm9D

    work page 2024

  46. [46]

    Wicker SB (1995) Error control systems for digital communication and storage, volume 1 (Prentice hall Englewood Cliffs)

    work page 1995

  47. [47]

    https://www.wipo.int/wipolex/en/legislation/details/21065, english version of the amended Copyright Law

    WIPO Lex (2020) Copyright law of the people's republic of china. https://www.wipo.int/wipolex/en/legislation/details/21065, english version of the amended Copyright Law

    work page 2020

  48. [48]

    World Intellectual Property Organization (2021) Digital date-and-time-stamping: The evidentiary value and practical significance of wipo proof

    work page 2021

  49. [49]

    International Conference on Machine Learning, 53443--53470 (PMLR)

    Wu Y, Hu Z, Guo J, Zhang H, Huang H (2024) A resilient and accessible distribution-preserving watermark for large language models. International Conference on Machine Learning, 53443--53470 (PMLR)

    work page 2024

  50. [50]

    Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), 4031--4055

    Yoo K, Ahn W, Kwak N (2024) Advancing beyond identification: Multi-bit watermark for large language models. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), 4031--4055

    work page 2024

  51. [51]

    33rd USENIX Security Symposium (USENIX Security 24), 1813--1830

    Zhang R, Hussain SS, Neekhara P, Koushanfar F (2024) \ REMARK-LLM \ : A robust and efficient watermarking framework for generative large language models. 33rd USENIX Security Symposium (USENIX Security 24), 1813--1830

    work page 2024