Pith. sign in

REVIEW 3 major objections 4 minor 78 references

The End Of Universal Lifelong Identifiers: Identity Systems For The AI Era

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper argues that Universal Lifelong Identifiers like SSN and Aadhaar are structurally unsafe in the AI era, and that a Merkle-commitment plus zero-knowledge framework can replace them without disrupting existing workflows.

desk verdict Valuable AI-era threat model, but the verification circuit is replayable and the drop-in replacement claim is not supported. read the letter →

arxiv 2506.02027 v1 pith:OSN2PRV7 submitted 2025-05-29 cs.CR cs.AIcs.CY

Shriphani Palakodety This is my paper
classification cs.CRcs.AIcs.CY
keywords universallifelongidentifiersprivacyzero-knowledgeproofsunlinkabilityidentitysystemsAIthreatmodelMerkletreesanonymouscredentials
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This position paper argues that Universal Lifelong Identifiers (ULIs)—single static numbers like the Social Security Number and Aadhaar, reused across health care, finance, and government—are structurally unsafe once AI systems can extract, memorize, and cross-link them. The author contends that redaction, consent, and access control cannot fix this because exposure becomes permanent once identifiers sink into model training data. The core positive claim is that a practical replacement exists: individuals privately generate many unlinkable identifiers, commit to them in a Merkle tree, and prove their legitimacy in zero knowledge against public allow and block lists. If this is right, institutions could keep their existing forms and workflows while the identifiers themselves stop functioning as universal join keys, making AI-driven linkage and surveillance much harder.

What carries the argument

The load-bearing mechanism is the zero-knowledge circuit VERIFYIDENTIFIER (Algorithm 1) paired with two public Merkle roots maintained by a coordinator: an allow root $A$ over valid identity commitments and a block root $B$, implemented as a sparse Merkle tree, over revoked commitments. The prover knows a witness containing an identifier $id$, a hidden commitment $I$, and inclusion and non-inclusion proofs, and demonstrates that $id$ is a leaf of some commitment $I$, that $I$ is in $A$, and that $I$ is not in $B$, without revealing $I$. This single circuit carries the argument: it lets identifiers be random, per-context strings that still look like ordinary alphanumeric IDs for forms, while the legitimacy check is cryptographic and unlinkable.

What would settle it

Run Algorithm 1 in the paper's own terms: take any identifier id that is a leaf of any valid, non-revoked identity commitment I, construct the witness without any authenticator or secret, and submit the resulting proof to a verifier with the current allow and block roots. If the verifier accepts, which the circuit's constraints permit, then possession of a single transcript is sufficient to impersonate the owner. Alternatively, capture one valid proof from a real interaction and replay it unchanged in a second interaction with the same roots; acceptance would confirm the missing holder binding.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that ULI failure is not an engineering bug but an architectural one: persistent cross-domain reuse makes any single leaked number a permanent join key, and commoditized AI tools have removed the practical barriers that once protected those numbers. Against this threat, the paper proposes changing the identifier itself rather than cleaning data or editing models after the fact. Each person holds a private portfolio of identifiers, commits to the portfolio with a Merkle root, registers only that root with a coordinator, and for each workflow presents one identifier plus a zero-knowledge proof that it belongs to a valid, unrevoked commitment. The proof hides which commitment, so no two services can tell they are dealing with the same person, while verifiers can still check legitimacy offline against public allow and block roots. The claimed result is comprehensive privacy, including forward, per-relying-party, and per-interaction unlinkability, with auditable delegation and minimal disruption to existing identifier workflows.

Load-bearing premise

The whole design rests on the assumption that a zero-knowledge proof of identifier legitimacy is enough to identify the person presenting it; the circuit as written proves only that the identifier belongs to some valid, unrevoked commitment, with no secret known only to the owner bound into the proof.

Editorial extensions

If this is right

  • A breach of one service's database would leak identifiers that are useless for linking that individual's records at any other service, because the same person presents different identifiers elsewhere.
  • Institutions could adopt the system without redesigning forms, databases, or document workflows: the presented identifier remains an ordinary alphanumeric string, and verification happens via a proof against public roots.
  • Regulators and auditors retain a real function: public allow and block roots make authorization and revocation auditable without revealing who used which identifier.
  • Delegation (power of attorney, caregivers) can be implemented by handing over Merkle subtrees, so the delegate proves legitimacy without learning the principal's other identifiers.
  • Migration can begin incrementally by deriving the new per-context identifiers deterministically from existing ULIs, allowing a gradual transition rather than a single cutover.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • [Inference] As written, the proof circuit does not bind the proof to a secret held by the presenter, so a captured proof transcript could be replayed by someone who is not the identifier's owner; the optional authenticator binding would need to be inside the circuit for the system to work as a secure drop-in replacement.
  • [Inference] Because deterministic derivation from an existing ULI is proposed as a migration seed, any future leak of that seed would re-link the entire portfolio; a safer migration would require the old ULI to be retired and destroyed rather than retained as a backup.
  • [Inference] The same architecture could plausibly extend to machine identities and IoT devices, where persistent device identifiers are routinely exposed in network metadata, but the coordinator and allow-list model would need to handle highly ephemeral and frequently compromised devices.
  • [Inference] The threat model implies that even a fully deployed system cannot satisfy regulators who require proof of identity at the exact instant of a transaction unless the verifier can somehow bind the proof to the physical presenter, a gap the paper leaves open.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper argues that Universal Lifelong Identifiers (ULIs) such as SSNs and Aadhaar are fundamentally incompatible with AI-era privacy and must be phased out in favor of unlinkable, cryptographically scoped identifiers. It presents a threat model centered on AI-driven extraction, memorization, and linkage; defines eight desired properties for identity systems; and sketches a framework in which users self-generate portfolios of identifiers committed in a Merkle tree, a coordinator maintains public allow/block roots, and zero-knowledge proofs demonstrate that a presented identifier belongs to a valid, non-revoked commitment. The paper claims this framework satisfies all eight properties while retaining workflow compatibility, and concludes that comprehensive privacy is achievable without sacrificing functionality.

Significance. The motivating analysis is valuable: the threat model is current and well-referenced, and the use of standard primitives (Merkle trees, sparse Merkle trees, zk-SNARKs) makes the architectural direction plausible. The paper is honestly framed as a conceptual/position contribution and includes a limitations section that acknowledges engineering and deployment challenges. However, the central claim is not supported as written: the verification circuit does not bind a proof to the presenter, so captured transcripts are replayable, and the property list in Section 6.5 is asserted without formal definitions or security arguments. The framework may be a useful starting point, but the conclusions currently overstate what has been demonstrated.

major comments (3)
  1. [6.4 (Algorithm 1)] The VERIFYIDENTIFIER circuit does not bind the proof to the presenter. The witness contains only id, the identity commitment I, and Merkle proofs; the public inputs are id, A, and B. There is no secret key, challenge nonce, or verifier identity in the circuit, and Section 6.4 explicitly excludes identity bindings and interaction-specific gadgets. Consequently, anyone who captures a transcript (id, π) can replay it to a new verifier and be accepted as the holder of a legitimate identifier. This breaks the authentication function that ULIs serve and undermines the drop-in replacement claim. The optional authenticator binding I' ← CHF(I ∥ authenticator) in Section 6.2 is not incorporated, so the current construction is a bearer-token system rather than an identity system. The circuit must include a presenter-binding mechanism (e.g., a proof of knowledge of a secret associated with the commitment, or a signature over a verifier-supplied challenge), and the claims in Sections 6.5 and 10 must be revised accordingly.
  2. [6.5] The assertion that the proposed system satisfies the eight properties of Section 4 is not supported by any formal definitions, security games, or proofs. Terms such as 'forward unlinkability,' 'per-relying-party unlinkability,' and 'relying party anonymity' are not defined with respect to an adversary model, so the claims are not falsifiable. At minimum, the paper should provide precise informal definitions of the adversarial capabilities and the unlinkability/anonymity notions, and argue why the construction meets them. As written, the property list in Section 6.5 is a design target, not a verified result.
  3. [8 and 10] Section 8 lists 'protocol formalization' as a limitation and future work, yet Section 10 concludes that 'we have shown that comprehensive privacy is achievable without sacrificing functionality through a cryptographic architecture.' This is an unsupported leap: if the protocol is not fully specified, the conclusion should be hedged to state that the framework suggests a path toward such a system, not that it has been demonstrated.
minor comments (4)
  1. [5.1 and 6.2] The term 'identity commitment' is used for a Merkle root of a set of identifiers, but the paper never states what binding or hiding properties are expected of this commitment. Clarify the commitment's formal properties and whether it is intended to be hiding against coordinators and verifiers.
  2. [6.2] The sentence 'Identifiers may also encode commitments to attributes for selective disclosure' is undeveloped. It is unclear how attribute commitments would be embedded in identifiers and how they would interact with the VERIFYIDENTIFIER circuit.
  3. [Algorithm 1] The presentation of Algorithm 1 lists id as both a witness and a public input. In a zk-SNARK, id is a public input and should not be part of the witness; the text should be adjusted to avoid confusion.
  4. [Figure 1] Figure 1 is referenced in Section 6.5 but is not described in the text. Add a sentence explaining what the figure shows and how the three components relate to the verification flow.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the framework is a conceptual construction whose unproven security claims are correctness concerns, not circular reasoning.

full rationale

The paper is a position paper that defines a threat model, derives eight desired properties for identifiers, and sketches a cryptographic framework built from standard primitives (Merkle trees, sparse Merkle trees, zk-SNARKs). The central claim—that the framework satisfies the properties and offers comprehensive privacy—is an assertion about a proposed design, not a derivation that reduces to its own inputs. There are no fitted parameters, no empirical predictions, and no normalization choices. The properties are motivated by the threat model and prior literature, not by the framework's outputs, so they are not self-definitional. The paper does not rely on citations to the author's own prior work; references are to established cryptographic and legal scholarship, so there is no self-citation load-bearing chain. The framework is honestly scoped: Section 5 states it is 'not intended as an exhaustive protocol specification, but rather as a demonstration of plausibility,' and Section 8 lists 'protocol formalization' as future work. The most serious weakness—Algorithm 1's VERIFYIDENTIFIER lacks holder binding, making non-interactive transcripts replayable—is a soundness/security gap in an incomplete protocol, not a circularity. The 'properties satisfied' statements in Section 6.5 are unproven claims about a sketch, but they are not derived from the properties themselves. No step in the paper equates a prediction to a fit, renames a known result while presenting it as new, or imports a uniqueness theorem from the authors' own prior work. Therefore the circularity score is 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 3 invented entities

The design relies on standard cryptographic assumptions plus a trusted coordinator and capable client devices. The invented components (identity commitments, allow/block roots, private portfolios) have no independent implementation or falsifiable evidence outside the paper.

assumptions (4)
  • standard math Collision resistance and preimage resistance of the cryptographic hash function, plus soundness and zero-knowledge of the zk-SNARK.
    Invoked in Section 6.1; if these fail, membership proofs could be forged or identifiers revealed.
  • domain assumption A trusted coordinator maintains allow and block roots, never learns underlying identifiers, and cannot link usage to individuals.
    Sections 5.2 and 6.3 assume this; a malicious or compromised coordinator could register commitments under known identities and correlate proofs, breaking relying-party anonymity and unlinkability.
  • domain assumption Participants can securely generate, store, and present private identifier portfolios and zero-knowledge witnesses on capable devices.
    Section 8 lists secret management and hardware requirements as limitations; if participants leak seeds or witnesses, adversaries can generate or replay valid proofs.
  • domain assumption Existing institutional workflows can accept alphanumeric identifiers plus offline zero-knowledge verification.
    Section 7 claims drop-in workflow compatibility, but Section 8 says the approach is incompatible with passive physical credentials and requires proof-generation hardware, so the compatibility property holds only for a subset of workflows.
invented entities (3)
  • Identity commitment (Merkle root of a participant's private identifier portfolio)
    purpose: Anchors all legitimacy proofs and coordinator registration without revealing individual identifiers
    The security of the commitment scheme is asserted, not proven, and no implementation is provided (Sections 5.2, 6.2).
  • Coordinator-maintained allow and block roots
    purpose: Publicly auditable authorization and revocation lists enabling offline verification
    The system depends on these roots, but their trust, update, and audit mechanisms are only sketched (Sections 5.1, 6.3).
  • Private identifier portfolio
    purpose: Provides per-interaction and per-relying-party identifiers to prevent cross-context linkage
    Generation, storage, and rotation are described at a high level; secret management is listed as an open limitation (Sections 6.2, 8).

how reviews work

0 comments
Cite this review

Pith. "Pith review of The End Of Universal Lifelong Identifiers: Identity Systems For The AI Era." pith.science (2026). https://pith.science/paper/OSN2PRV7

@misc{pith2026250602027,
  author       = {Pith},
  title        = {Pith review of: The End Of Universal Lifelong Identifiers: Identity Systems For The AI Era},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OSN2PRV7}},
  note         = {Machine review of arXiv:2506.02027}
}
read the original abstract

Many identity systems assign a single, static identifier to an individual for life, reused across domains like healthcare, finance, and education. These Universal Lifelong Identifiers (ULIs) underpin critical workflows but now pose systemic privacy risks. We take the position that ULIs are fundamentally incompatible with the AI era and must be phased out. We articulate a threat model grounded in modern AI capabilities and show that traditional safeguards such as redaction, consent, and access controls are no longer sufficient. We define core properties for identity systems in the AI era and present a cryptographic framework that satisfies them while retaining compatibility with existing identifier workflows. Our design preserves institutional workflows, supports essential functions such as auditability and delegation, and offers a practical migration path beyond ULIs.

Figures

Figures reproduced from arXiv: 2506.02027 by the authors.

Figure 1
Figure 1. Overview of our unlinkable identifier framework showing the three key components: [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

78 extracted references · 57 canonical work pages

  1. [1]

    URL https://www.europol

    Internet organised crime threat assessment (iocta), 2020. URL https://www.europol. europa.eu. Accessed: 2025-05-21

    work page 2020

  2. [2]

    Mohammad Abomhara and Geir M. Koien. Cyber security and the internet of things: Vulnerabilities, threats, intruders and attacks.Journal of Big Data, 2015. URL https: //journalofbigdata.springeropen.com/articles/10.1186/s40537-016-0059-y

    work page doi:10.1186/s40537-016-0059-y 2015

  3. [3]

    Pii masker: Transformer-based entity masking library

    HydroX AI. Pii masker: Transformer-based entity masking library. https://github.com/ HydroXai/pii-masker, 2024. Accessed: 2025-05-21. 9

    work page 2024

  4. [4]

    The path to self-sovereign identity, 2016

    Christopher Allen. The path to self-sovereign identity, 2016. URL http://www. lifewithalacrity.com/2016/04/the-path-to-self-soverereign-identity. html. Blog post

    work page 2016

  5. [5]

    Irma: practical, decentralized and privacy-friendly identity management using smartphones

    Gergely Alpár, Fabian Van Den Broek, Brinda Hampiholi, Bart Jacobs, Wouter Lueks, and Sietse Ringers. Irma: practical, decentralized and privacy-friendly identity management using smartphones. In10th Workshop on Hot Topics in Privacy Enhancing Technologies (HotPETs 2017), pages 1–2, 2017

    work page 2017

  6. [6]

    Antonopoulos et al

    Andreas M. Antonopoulos et al. Bip-0032: Hierarchical deterministic wallets. https:// github.com/bitcoin/bips/blob/master/bip-0032.mediawiki, 2012

    work page 2012

  7. [7]

    The mighty torr: A benchmark for table reasoning and robustness, 2025

    Shir Ashury-Tahan, Yifan Mai, Rajmohan C, Ariel Gera, Yotam Perlitz, Asaf Yehudai, Elron Bandel, Leshem Choshen, Eyal Shnarch, Percy Liang, and Michal Shmueli-Scheuer. The mighty torr: A benchmark for table reasoning and robustness, 2025. URL https://arxiv. org/abs/2502.19412

    arXiv 2025

  8. [8]

    Domain-specific pseudony- mous signatures for the german identity card

    Jens Bender, Özgür Dagdelen, Marc Fischlin, and Dennis Kügler. Domain-specific pseudony- mous signatures for the german identity card. In Dieter Gollmann and Felix C. Freiling, editors, Information Security, pages 104–119, Berlin, Heidelberg, 2012. Springer Berlin Heidelberg. ISBN 978-3-642-33383-5

    work page 2012

Show all 78 references
  1. [9]

    An efficient system for non-transferable anonymous credentials with optional anonymity revocation

    Jan Camenisch and Anna Lysyanskaya. An efficient system for non-transferable anonymous credentials with optional anonymity revocation. In Birgit Pfitzmann, editor,Advances in Cryptology — EUROCRYPT 2001, volume 2045 ofLecture Notes in Computer Science, pages 93–118. Springer, ...

    2001 doi

  2. [10]

    The secret sharer: Evaluating and testing unintended memorization in neural networks

    Nicholas Carlini, Chang Liu, Úlfar Erlingsson, Jernej Kos, and Dawn Song. The secret sharer: Evaluating and testing unintended memorization in neural networks. In28th USENIX security symposium (USENIX security 19), pages 267–284, 2019

    2019

  3. [11]

    Extracting training data from large language models

    Nicholas Carlini, Florian Tramèr, Eric Wallace, Matthew Jagielski, Ariel Herbert-V oss, Kather- ine Lee, Adam Roberts, Tom Brown, Dawn Song, Úlfar Erlingsson, Alina Oprea, and Colin Raffel. Extracting training data from large language models. In30th USENIX Security Sym- posium...

    2021

  4. [12]

    The signal private group system and anony- mous credentials supporting efficient verifiable encryption

    Melissa Chase, Trevor Perrin, and Greg Zaverucha. The signal private group system and anony- mous credentials supporting efficient verifiable encryption. InProceedings of the 2020 ACM SIGSAC Conference on Computer and Communications Security, CCS ’20, page 1445–1459, New York,...

    2020

  5. [13]

    What is the purpose of having a social security number (ssn)?,

    James Chen. What is the purpose of having a social security number (ssn)?,

  6. [14]

    Yale University Press, 2012

    Julie E Cohen.Configuring the networked self: Law, code, and the play of everyday practice. Yale University Press, 2012

    2012

  7. [15]

    See 6 charts that show the astonishing rise of artificial intelligence.TIME, Jul

    Ken Dilanian. See 6 charts that show the astonishing rise of artificial intelligence.TIME, Jul

  8. [16]

    How unique is your web browser? InProceedings of the 10th International Conference on Privacy Enhancing Technologies, PETS’10, page 1–18, Berlin, Heidelberg, 2010

    Peter Eckersley. How unique is your web browser? InProceedings of the 10th International Conference on Privacy Enhancing Technologies, PETS’10, page 1–18, Berlin, Heidelberg, 2010. Springer-Verlag. ISBN 3642145264

    2010

  9. [17]

    Accessed: 2025-05-12

    URL https://time.com/6300942/ai-progress-charts/ . Accessed: 2025-05-12

    2025

  10. [18]

    Succinct non-interactive zero knowledge for a von neumann architecture

    Rosario Gennaro, Craig Gentry, Bryan Parno, and Mariana Raykova. Succinct non-interactive zero knowledge for a von neumann architecture. InCRYPTO, 2013

    2013

  11. [19]

    Ai risks: Optical character recogni- tion

    European Data Protection Board. Ai risks: Optical character recogni- tion. https://www.edpb.europa.eu/system/files/2024-06/ai-risks_ d2optical-character-recognition_edpb-spe-programme_en_2.pdf , June 2024. Accessed: 2025-04-22

    2024

  12. [20]

    The knowledge complexity of interactive proof systems

    Shafi Goldwasser, Silvio Micali, and Charles Rackoff. The knowledge complexity of interactive proof systems. InSTOC, 1985

    1985

  13. [21]

    Lecture notes on cryptography, 2008

    Shafi Goldwasser and Mihir Bellare. Lecture notes on cryptography, 2008. Available at https://cseweb.ucsd.edu/~mihir/papers/gb.html. 10

    2008

  14. [22]

    On the size of pairing-based non-interactive arguments

    Jens Groth. On the size of pairing-based non-interactive arguments. InEUROCRYPT, 2016

    2016

  15. [23]

    Employers and the social insurance number (sin), 2023

    Government of Canada. Employers and the social insurance number (sin), 2023. URL https://www.canada.ca/en/employment-social-development/programs/ei/ ei-list/ei-employers-sin.html. Accessed April 30, 2025

    2023

  16. [24]

    Privacy-enhancing identity management

    Marit Hansen, Stefan Schwartz, and Alissa Cooper. Privacy-enhancing identity management. Information Security Technical Report, 13(4):181–186, 2008

    2008

  17. [25]

    Evaluating the usability of privacy choice mechanisms

    Hana Habib and Lorrie Faith Cranor. Evaluating the usability of privacy choice mechanisms. InEighteenth Symposium on Usable Privacy and Security (SOUPS 2022), pages 273–289, Boston, MA, August 2022. USENIX Association. ISBN 978-1-939133-30-4. URL https: //www.usenix.org/confer...

    2022

  18. [26]

    Preventing generation of verbatim memorization in language models gives a false sense of privacy

    Daphne Ippolito, Florian Tramer, Milad Nasr, Chiyuan Zhang, Matthew Jagielski, Katherine Lee, Christopher Choquette Choo, and Nicholas Carlini. Preventing generation of verbatim memorization in language models gives a false sense of privacy. In C. Maria Keet, Hung-Yi Lee, and ...

    2023 doi

  19. [27]

    Personal data breaches, 2023

    Information Commissioner’s Office (ICO). Personal data breaches, 2023. URL https: //ico.org.uk/for-organisations/report-a-breach/personal-data-breach/ personal-data-breaches-a-guide/

    2023

  20. [28]

    Unique health identifiers for universal health coverage.Bulletin of the World Health Organization, 97(10):674–680, 2019

    Juhwan Kim, Jong-Koo Lee, and Hyun-Sun Kim. Unique health identifiers for universal health coverage.Bulletin of the World Health Organization, 97(10):674–680, 2019. doi: 10.2471/BLT.18.226514

    2019 doi

  21. [29]

    A systematic analysis of the capital one data breach: Critical lessons learned.ACM Trans

    Shaharyar Khan, Ilya Kabanov, Yunke Hua, and Stuart Madnick. A systematic analysis of the capital one data breach: Critical lessons learned.ACM Trans. Priv. Secur., 26(1), November

  22. [30]

    HMAC-based Extract-and-Expand Key Derivation Function (HKDF)

    Hugo Krawczyk and Pasi Eronen. HMAC-based Extract-and-Expand Key Derivation Function (HKDF). RFC 5869, May 2010. URLhttps://www.rfc-editor.org/info/rfc5869

    2010

  23. [31]

    Revocation transparency

    Ben Laurie and Emilia Kasper. Revocation transparency. https://www.links.org/files/ RevocationTransparency.pdf, 2012. Accessed: 2025-05-14

    2012

  24. [32]

    Propile: probing privacy leakage in large language models

    Siwon Kim, Sangdoo Yun, Hwaran Lee, Martin Gubri, Sungroh Yoon, and Seong Joon Oh. Propile: probing privacy leakage in large language models. InProceedings of the 37th Interna- tional Conference on Neural Information Processing Systems, NIPS ’23, Red Hook, NY , USA,

  25. [33]

    Curran Associates Inc

  26. [34]

    VHELM: A holistic evaluation of vision language models

    Tony Lee, Haoqin Tu, Chi Heem Wong, Wenhao Zheng, Yiyang Zhou, Yifan Mai, Jos- selin Somerville Roberts, Michihiro Yasunaga, Huaxiu Yao, Cihang Xie, and Percy Liang. VHELM: A holistic evaluation of vision language models. InThe Thirty-eight Conference on Neural Information Pro...

    2024

  27. [35]

    Trocr: transformer-based optical character recognition with pre- trained models

    Minghao Li, Tengchao Lv, Jingye Chen, Lei Cui, Yijuan Lu, Dinei Florencio, Cha Zhang, Zhoujun Li, and Furu Wei. Trocr: transformer-based optical character recognition with pre- trained models. InProceedings of the Thirty-Seventh AAAI Conference on Artificial Intelligence and T...

    2023

  28. [36]

    A universally unique identifier (uuid) urn namespace.https://datatracker.ietf.org/doc/html/rfc4122, 2005

    Paul Leach, Michael Mealling, and Rich Salz. A universally unique identifier (uuid) urn namespace.https://datatracker.ietf.org/doc/html/rfc4122, 2005. RFC 4122

    2005

  29. [37]

    scrubadub: Automatically clean personally identifiable information (pii) from dirty dirty text.https://github.com/LeapBeyond/scrubadub, 2018

    Sam Lee. scrubadub: Automatically clean personally identifiable information (pii) from dirty dirty text.https://github.com/LeapBeyond/scrubadub, 2018. Version 2.0+

    2018

  30. [38]

    Assessing poor adoption of the eid in germany

    Philipp Liesbrock and Eriks Sneiders. Assessing poor adoption of the eid in germany. In Alvaro Rocha, Hojjat Adeli, Gintautas Dzemyda, Fernando Moreira, and Valentina Colla, editors,Information Systems and Technologies, pages 292–301, Cham, 2024. Springer Nature Switzerland. I...

    2024

  31. [39]

    eidas implementation challenges: The case of estonia and the netherlands

    Silvia Lips, Nitesh Bharosa, and Dirk Draheim. eidas implementation challenges: The case of estonia and the netherlands. In Andrei Chugunov, Igor Khodachek, Yuri Misnikov, and Dmitrii Trutnev, editors,Electronic Governance and Open Society: Challenges in Eurasia, pages 75–89, ...

    2020

  32. [40]

    Htr-vt: Handwritten text recognition with vision transformer.Pattern Recognition, 158:110967, 2025

    Yuting Li, Dexiong Chen, Tinglong Tang, and Xi Shen. Htr-vt: Handwritten text recognition with vision transformer.Pattern Recognition, 158:110967, 2025. ISSN 0031-3203. doi: https://doi.org/10.1016/j.patcog.2024.110967. URL https://www.sciencedirect.com/ science/article/pii/S0...

    2025

  33. [41]

    Percy Liang, Rishi Bommasani, Tony Lee, Dimitris Tsipras, Dilara Soylu, Michihiro Yasunaga, Yian Zhang, Deepak Narayanan, Yuhuai Wu, Ananya Kumar, Benjamin Newman, Binhang Yuan, Bobby Yan, Ce Zhang, Christian Alexander Cosgrove, Christopher D Manning, Christopher Re, Diana Aco...

    2023

  34. [42]

    Mass editing memory in a transformer.The Eleventh International Conference on Learning Represen- tations (ICLR), 2023

    Kevin Meng, Arnab Sen Sharma, Alex Andonian, Yonatan Belinkov, and David Bau. Mass editing memory in a transformer.The Eleventh International Conference on Learning Represen- tations (ICLR), 2023

    2023

  35. [43]

    Protocols for public key cryptosystems

    Ralph C Merkle. Protocols for public key cryptosystems. InIEEE Symposium on Security and Privacy, pages 122–134, 1980

    1980

  36. [44]

    Analyzing Leakage of Personally Identifiable Information in Language Models

    Nils Lukas, Ahmed Salem, Robert Sim, Shruti Tople, Lukas Wutschitz, and Santiago Zanella- Beguelin. Analyzing Leakage of Personally Identifiable Information in Language Models . In2023 IEEE Symposium on Security and Privacy (SP), pages 346–363, Los Alamitos, CA, USA, May 2023....

    2023

  37. [45]

    Locating and editing factual associations in gpt

    Kevin Meng, David Bau, Alex Andonian, and Yonatan Belinkov. Locating and editing factual associations in gpt. InProceedings of the 36th International Conference on Neural Information Processing Systems, NIPS ’22, Red Hook, NY , USA, 2022. Curran Associates Inc. ISBN 9781713871088

    2022

  38. [46]

    Stanford University Press, 2009

    Helen Nissenbaum.Privacy in context: Technology, policy, and the integrity of social life. Stanford University Press, 2009

    2009

  39. [47]

    Nemo curator: Pii identification and removal

    NVIDIA. Nemo curator: Pii identification and removal. https:// docs.nvidia.com/nemo-framework/user-guide/latest/datacuration/ personalidentifiableinformationidentificationandremoval.html, 2024. Ac- cessed: 2025-05-21

    2024

  40. [48]

    Robust de-anonymization of large sparse datasets

    Arvind Narayanan and Vitaly Shmatikov. Robust de-anonymization of large sparse datasets. In2008 IEEE Symposium on Security and Privacy (sp 2008), pages 111–125, 2008. doi: 10.1109/SP.2008.33

    2008 doi

  41. [49]

    Cookieless monster: Exploring the ecosystem of web-based device fingerprinting

    Nick Nikiforakis, Alexandros Kapravelos, Wouter Joosen, Christopher Kruegel, Frank Piessens, and Giovanni Vigna. Cookieless monster: Exploring the ecosystem of web-based device fingerprinting. In2013 IEEE Symposium on Security and Privacy, pages 541–555, 2013. doi: 10.1109/SP.2013.43

    2013 doi

  42. [50]

    Training language models to follow instructions with human feedback.Advances in neural information processing systems, 35:27730–27744, 2022

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback.Advances in neural information processing systems, 35:27730...

    2022

  43. [51]

    Choquette-Choo, Zhengming Zhang, Yaoqing Yang, and Prateek Mittal

    Ashwinee Panda, Christopher A. Choquette-Choo, Zhengming Zhang, Yaoqing Yang, and Prateek Mittal. Teach LLMs to phish: Stealing private information from language models. InThe Twelfth International Conference on Learning Representations, 2024. URL https: //openreview.net/forum...

    2024

  44. [52]

    Virtual id (vid)

    Unique Identification Authority of India. Virtual id (vid). https://uidai.gov.in/ my-aadhaar/about-your-aadhaar/virtual-id.html, 2023. Accessed: 2025-05-02

    2023

  45. [53]

    Broken promises of privacy: Responding to the surprising failure of anonymization

    Paul Ohm. Broken promises of privacy: Responding to the surprising failure of anonymization. UCLA L. Rev., 57:1701, 2010. 12

    2010

  46. [54]

    Learn to unlearn: Insights into machine unlearning.Computer, 57(3):79–90, 2024

    Youyang Qu, Xin Yuan, Ming Ding, Wei Ni, Thierry Rakotoarivelo, and David Smith. Learn to unlearn: Insights into machine unlearning.Computer, 57(3):79–90, 2024. doi: 10.1109/MC. 2023.3333319

    2024

  47. [55]

    50 ways to leak your data: An exploration of apps’ circumvention of the android permissions system

    Joel Reardon, Álvaro Feal, Primal Wijesekera, Amit Elazari Bar On, Narseo Vallina-Rodriguez, and Serge Egelman. 50 ways to leak your data: An exploration of apps’ circumvention of the android permissions system. In28th USENIX Security Symposium (USENIX Security 19), pages 603–...

    2019

  48. [56]

    U-prove cryptographic specification v1.1

    Christian Paquin and Greg Zaverucha. U-prove cryptographic specification v1.1. InMicrosoft Corporation, 2011

    2011

  49. [57]

    The story of the social security number.Social Security Bulletin, 69(2): 55–64, 2009

    Carolyn Puckett. The story of the social security number.Social Security Bulletin, 69(2): 55–64, 2009. URL https://www.ssa.gov/policy/docs/ssb/v69n2/v69n2p55.html. Accessed: 2025-04-21

    2009

  50. [58]

    Identifying and mitigating privacy risks stemming from language models: A survey.ArXiv, abs/2310.01424,

    Victoria Smith, Ali Shahin Shamsabadi, Carolyn Ashurst, and Adrian Weller. Identifying and mitigating privacy risks stemming from language models: A survey.ArXiv, abs/2310.01424,

    arXiv

  51. [59]

    Daniel J. Solove. A taxonomy of privacy.University of Pennsylvania Law Review, 154(3): 477–560, 2006

    2006

  52. [60]

    University of North Carolina Press, 1995

    Priscilla M Regan.Legislating privacy: Technology, social values, and public policy. University of North Carolina Press, 1995

    1995

  53. [61]

    Estimating the success of re-identifications in incomplete datasets using generative models.Nature Communications, 10, 07 2019

    Luc Rocher, Julien Hendrickx, and Yves-Alexandre Montjoye. Estimating the success of re-identifications in incomplete datasets using generative models.Nature Communications, 10, 07 2019. doi: 10.1038/s41467-019-10933-3

    2019 doi

  54. [62]

    Chatgpt answer goes wrong, gives away journalist’s number to join signal, 2023

    LiveMint Staff. Chatgpt answer goes wrong, gives away journalist’s number to join signal, 2023. URL https://www.livemint.com/news/ chatgpt-answer-goes-wrong-gives-away-journalist-s-number-to-join-signal-11676625029542. html. Accessed: 2025-05-02

    2023

  55. [63]

    URLhttps://api.semanticscholar.org/CorpusID:263608702

  56. [64]

    About aadhaar, 2024

    Unique Identification Authority of India. About aadhaar, 2024. URL https://uidai.gov. in/en/?option=com_content&view=article&id=14. Accessed: 2025-04-21

    2024

  57. [65]

    Srinivasan, Q

    S. Srinivasan, Q. Pitcher, and J.S. Goldberg.Data Breach at Equifax. Main Case. Harvard Business School, 2019. URLhttps://books.google.com/books?id=LmVCzgEACAAJ

    2019

  58. [66]

    Beyond memorization: Violating privacy via inference with large language models

    Robin Staab, Mark Vero, Mislav Balunovi ´c, and Martin Vechev. Beyond memorization: Violating privacy via inference with large language models. InThe Twelfth International Conference on Learning Representations, 2024

    2024

  59. [67]

    Department of Justice

    U.S. Department of Justice. Overview of the privacy act of 1974, 2020. URL https://www.justice.gov/opcl/overview-privacy-act-1974-2020-edition/ disclosures-third-parties. 13

    1974

  60. [68]

    Simple demographics often identify people uniquely.Health (San Francisco), 671:1–34, 2000

    Latanya Sweeney. Simple demographics often identify people uniquely.Health (San Francisco), 671:1–34, 2000

    2000

  61. [69]

    The right to privacy.Harvard Law Review, 4(5): 193–220, 1890

    Samuel D Warren and Louis D Brandeis. The right to privacy.Harvard Law Review, 4(5): 193–220, 1890

  62. [70]

    Social security numbers: Federal and state laws restrict use of ssns, yet gaps remain

    United States Government Accountability Office. Social security numbers: Federal and state laws restrict use of ssns, yet gaps remain. Technical Report GAO-05-1016T, U.S. Government Accountability Office, 2005. URL https://www.gao.gov/products/gao-05-1016t . Tes- timony Before...

    2005

  63. [71]

    Congress

    U.S. Congress. Health insurance portability and accountability act of 1996. https://www. govinfo.gov/app/details/PLAW-104publ191, 1996. Public Law 104–191

    1996

  64. [72]

    Quantifying and analyzing entity-level memorization in large language models

    Zhenhong Zhou, Jiuyang Xiang, Chaomeng Chen, and Sen Su. Quantifying and analyzing entity-level memorization in large language models. InProceedings of the Thirty-Eighth AAAI Conference on Artificial Intelligence and Thirty-Sixth Conference on Innovative Applications of Artifi...

    2024 doi

  65. [73]

    Department of the Treasury, Office of Foreign Assets Control

    U.S. Department of the Treasury, Office of Foreign Assets Control. A framework for ofac compli- ance commitments, 2019. URL https://ofac.treasury.gov/media/16331/download? inline. Accessed: 2025-05-17

    2019

  66. [75]

    Heng Xu, Tianqing Zhu, Lefeng Zhang, Wanlei Zhou, and Philip S. Yu. Machine unlearning: A survey.ACM Comput. Surv., 56(1), August 2023. ISSN 0360-0300. doi: 10.1145/3603620. URLhttps://doi.org/10.1145/3603620

    2023 doi

  67. [76]

    Privacy-preserving instructions for aligning large language models

    Da Yu, Peter Kairouz, Sewoong Oh, and Zheng Xu. Privacy-preserving instructions for aligning large language models. InProceedings of the 41st International Conference on Machine Learning, ICML’24. JMLR.org, 2024

    2024

  68. [78]

    Examining the adoption and abandonment of security, privacy, and identity theft protection practices

    Yixin Zou, Kevin Roundy, Acar Tamersoy, Saurabh Shintre, Johann Roturier, and Flo- rian Schaub. Examining the adoption and abandonment of security, privacy, and identity theft protection practices. InProceedings of the 2020 CHI Conference on Human Fac- tors in Computing System...

    2020

  69. [2022]

    doi: 10.1145/3546068

    ISSN 2471-2566. doi: 10.1145/3546068. URL https://doi.org/10.1145/3546068

    doi

  70. [2023]

    Accessed: 2025-04-22

    URL https://www.investopedia.com/articles/personal-finance/050615/ purpose-having-social-security-number.asp. Accessed: 2025-04-22

    2025

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.