Pith. sign in

REVIEW 3 major objections 6 minor 154 references

Deepfake research targets fake images, not the people harmed by AI-generated sexual abuse.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 13:46 UTC pith:WNLEB4CW

load-bearing objection Conceptually strong position paper, but the '0 of 39' empirical claim is an artifact of a search string that pre-selects authenticity-oriented papers. the 3 major comments →

arxiv 2607.18263 v2 pith:WNLEB4CW submitted 2026-05-16 cs.AI

Position: AI/ML Deepfake Research is Misaligned with AI-Generated Non-Consensual Intimate Imagery (AIG-NCII)

Li Qiwei , Wells Lucas Santo , Sarita Schoenebeck , Eric Gilbert This is my paper
classification cs.AI
keywords deepfakesAI-generated non-consensual intimate imageryAIG-NCIIdignity harmsauthenticity detectioncontent provenanceAI safetygenerative AI abuse
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This position paper argues that the AI/ML field's approach to deepfakes is misaligned with the actual dominant harm. The paper claims that technical interventions—detection, provenance, and watermarking—treat deepfakes as a threat to truth and authenticity, while the most common real-world abuse is AI-generated non-consensual intimate imagery (AIG-NCII), a dignity violation against the person depicted. A landscape analysis of 39 highly cited papers found zero technical implementations with a threat model specific to AIG-NCII. The core claim is that artificiality and consent are orthogonal axes of harm, so knowing an image is synthetic does not protect the subject and can actually worsen the harm. The paper calls for subject-centric threat models, harm-reduction metrics, and partnerships with experts in sexual violence prevention.

Core claim

The paper establishes that the AI/ML community's deepfake defenses rest on a category error: treating authenticity as a proxy for safety. Drawing on the human right to dignity, it distinguishes viewer-centric epistemic harms (fraud, deception, erosion of shared reality) from subject-centric dignity harms (non-consensual identity preservation and non-consensual modification, such as nudification). Through a qualitative analysis of 39 highly cited papers published in top venues between 2020 and 2025, it shows that the literature is almost entirely devoted to synthetic-versus-authentic discrimination, with only five papers mentioning sexual abuse in passing and zero papers building a technical

What carries the argument

The organizing device is the orthogonality claim (the paper's Table 1): the axis of artificiality (synthetic vs. authentic) is perpendicular to the axis of consent (safe vs. harmful), which means 'authentic ≠ safe' and that detection/provenance/watermarking tools measure the wrong axis. The framework is carried by a taxonomy of harms—viewer-centric epistemic harms versus subject-centric dignity harms—and by a landscape-analysis method that classifies papers into three tiers of engagement (no mention, mention only, technical implementation). AIG-NCII is defined as sexualized synthetic imagery of a specific real person created without consent, including 'nudification' via diffusion models.

Load-bearing premise

The paper's central claim hinges on the landscape analysis: if the 39-paper corpus, selected by authenticity-oriented search terms and top general-AI venues, is not representative of the broader research ecosystem, then the finding that zero technical interventions target AIG-NCII may be an artifact of the search rather than a fact about the field.

What would settle it

A landscape analysis that searches security and human-computer-interaction venues using the field's own vocabulary—'nudification', 'NCII', 'non-consensual intimate imagery', 'image-based sexual abuse'—would surface papers with AIG-NCII-specific threat models. Finding even a small number of such papers would disprove the 'zero technical implementations' result, though not necessarily the larger point about the field's dominant focus.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Public labeling of suspected AIG-NCII should be replaced by backend flags that trigger suppression or triage, treating synthetic abuse like traditional NCII.
  • Research goals should shift from maximizing detection accuracy to minimizing identity preservation, making it harder for a few reference photos to be cloned into sexualized content.
  • Threat models should incorporate adversaries who know the victim personally, as in intimate-partner violence, rather than assuming a stranger fraudster.
  • Safety metrics for generative AI should be validated by measured reductions in abuse prevalence, not by benchmark accuracy alone.
  • Release norms for high-fidelity identity-preserving and inpainting models should be revisited, with gated or researcher-only access.

Where Pith is reading between the lines

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

  • If the orthogonality claim is right, then adding synthetic-authentic labels to intimate imagery is not a neutral act; a direct test would compare victim-survivor outcomes on platforms that label-and-keep versus platforms that remove on suspicion.
  • The landscape analysis likely undercounts AIG-NCII-specific technical work, since its search terms ('detection', 'watermark', etc.) and venue filter exclude security and HCI venues where the paper itself cites relevant studies; a broader search would yield a more accurate denominator.
  • The argument implies that today's deepfake detection benchmarks, built on accuracy against synthetic-image datasets, may optimize for the wrong objective; a testable replacement is a benchmark that measures a model's ability to suppress generation of a specific identity after fine-tuning.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. This position paper argues that AI/ML deepfake research is structurally misaligned with the dominant real-world abuse of AI-generated non-consensual intimate imagery (AIG-NCII). The authors claim that current technical interventions focus on viewer-centric epistemic harms—authenticity, fraud, misinformation—while ignoring subject-centric dignity harms such as AIG-NCII. They support this with a landscape analysis of 39 highly-cited papers (2020–2025) that finds zero AIG-NCII-specific technical implementations, review detection/provenance/watermarking as the three dominant intervention paradigms, introduce an orthogonality framework distinguishing artificiality from consent (Table 1), and discuss ways in which authenticity tools may exacerbate dignity harms. The paper concludes with nine recommendations for realigning the field and a section addressing three alternative views.

Significance. If the empirical claim is supported, this paper makes an important conceptual contribution by reframing deepfake research and AI safety around consent and dignity rather than truth and authenticity. The orthogonality of artificiality and consent in Table 1 is a simple, useful corrective to the field's default assumption that synthetic detection equals safety. The paper is also methodologically honest: it concedes in §5 that definitive solutions cannot always be proposed, and in AV2 that adversarial defenses are brittle. The recommendations are concrete and ethically cautious, and the paper explicitly calls for partnerships with sexual-violence prevention experts. However, the central descriptive claim—that technical interventions 'almost entirely ignore' AIG-NCII—is currently not established by the landscape analysis, because the search and venue filters predetermine an authenticity-focused corpus. The conceptual argument is valuable, but the empirical framing must be revised or properly scoped.

major comments (3)
  1. [§2.2 and Table 2] The landscape analysis's Google Scholar query uses only authenticity-verification terms ('detection', 'detector', 'forensics', 'recognition', 'watermark') and excludes harm-related terms such as 'non-consensual', 'NCII', 'intimate imagery', 'nudify', or 'consent'. The venue filter (CVPR/ICCV/ECCV/NeurIPS/ICML/ICLR plus high-citation outliers) also excludes security venues. Yet the paper itself cites Gibson et al. (2025) and Han et al. (2025), both USENIX Security 25 papers on the AIG-NCII ecosystem, and Van Le et al. (2023), an ICCV paper on Anti-Dreambooth, as substantive AIG-NCII-related or subject-centric work; none appear in Table 2. Consequently, the '0 technical implementations' result and the abstract's claim that the research ecosystem is 'limited to authenticity detection tools' are at least partly artifacts of the search design. Please re-run the search with harm-specific terms
  2. [§2.2 and Figure 1] The classification of the 39 papers into 'No mention', 'Mention only', and 'Technical implementation' is presented without a coding protocol, inter-rater reliability, or a definition of 'meaningfully engaged'. The paper's headline result is the zero count in the last category, but this depends on a subjective threshold. Making the coding reproducible—for example, by including a coding sheet, specifying decision rules, and reporting dual-coding—would substantially strengthen the empirical claim.
  3. [Abstract and §5 R4] The paper's own recommendations and literature review cite subject-centric technical defenses—Anti-Dreambooth (Van Le et al., 2023), Glaze (Shan et al., 2023), AdvPaint (Jeon et al., 2025), and BlurGuard (Kim et al., 2026)—that are not detection, provenance, or watermarking. This is in tension with the abstract's statement that the research ecosystem is 'limited to authenticity detection tools.' Either these works were excluded from the landscape analysis because of the query/venue filters (reinforcing the issue in the first major comment), or they are considered out of scope, in which case the paper should say so explicitly. As written, the paper both asserts a monolithic authenticity-only ecosystem and cites counterexamples to that assertion.
minor comments (6)
  1. [§2.2] Table 2 includes several arXiv preprints (e.g., Liu et al., 2023; Ma et al., 2023) despite the stated top-venue filter. Clarify the inclusion rule for preprints and high-citation outliers.
  2. [References] The reference key '(tel, 2024)' appears in text as the author and should be formatted properly (The New York Times piece by Conger et al. or the Telegram article).
  3. [§2.2] 'This initial search criteria yielded 965 papers' — 'criteria' is plural; use 'criterion' or rephrase.
  4. [Figure 1] The figure caption states 'No papers found in our landscape analysis meaningfully engaged with AIG-NCII,' but the bar chart itself is not shown in the manuscript text; ensure the figure is legible and the axes are labeled.
  5. [§4.2 / Table 1] Consider replacing 'Consensual pornography' with 'Consensual intimate imagery' to avoid normative overtones and to align with the terminology used elsewhere in the paper.
  6. [References] Several references are missing venue or page information (e.g., Rini, 2020; Diel et al., 2025). Complete these for reproducibility.

Circularity Check

1 steps flagged

Landscape analysis partially constructs its own finding: the search and coding scheme are built from the authenticity paradigm, so the '0 AIG-NCII implementations' result is partly an artifact of the inclusion filter.

specific steps
  1. self definitional [§2.2 Methodology and Results; cf. §4.1]
    "We queried Google Scholar for papers containing a specific set of keywords, using the following query: (“detection” OR “detector” OR “forensics” OR “recognition” OR “watermark”) AND (“deepfake” OR “synthetic image” OR “fake image” OR “diffusion”). ... 3. Technical implementation (0 papers): The authors design an intervention with a threat model specific to AIG-NCII."

    The corpus is defined by authenticity-verification keywords (detection/detector/forensics/recognition/watermark), so the survey's finding that the community 'has coalesced around' detection, provenance, and watermarking and that 0 papers implement AIG-NCII-specific defenses largely recapitulates the inclusion filter. A subject-centric defense need not contain any query keyword, and the paper itself, in §4.1, identifies Anti-Dreambooth (Van Le et al., 2023) as a defense against the AIG-NCII harm mode of non-consensual identity preservation; that paper is absent from Table 2. The coding of 'technical implementation' is also term-based (NCII, revenge porn, nudity, undress, etc.), so the conclusion that no AIG-NCII-specific intervention exists is an artifact of the search/coding definition rat

full rationale

The central normative argument—that authenticity (synthetic/authentic) is orthogonal to consent (safe/harmful), so labels do not remediate dignity harms—does not reduce to any fitted parameter or self-citation; it is a conceptual argument supported by external reports, the Meta Oversight Board, and domain literature. The paper's self-citations (Qiwei et al. 2025 on DMCA inefficacy; Schoenebeck et al. 2021 on restorative justice) are contextual and not load-bearing. However, the paper's headline empirical claim, that deepfake technical research 'almost entirely ignores' AIG-NCII, rests on a landscape analysis whose query terms are themselves the authenticity paradigm. That result is thereby partly constructed by the search and coding scheme; the paper even cites subject-centric defenses in §4.1/R4 that would not enter the corpus. I therefore score partial circularity (6), not full circularity, since the paper's recommendations and the authentic≠safe framework retain independent content.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 0 invented entities

No numeric free parameters. The central claim rests on empirical prevalence assumptions and normative premises about consent and dignity, not on fitted quantities.

axioms (4)
  • domain assumption The majority of generative AI usage is AIG-NCII
    Supported by third-party reports about a single platform and period (Bouchaud 2026; CCDH 2026; Security Hero 2023); not independently verified in the paper.
  • domain assumption Dignity is a fundamental human right and subject-centric dignity harms are a valid criterion for evaluating technical interventions
    Adopted from UN UDHR and Citron (2018); a normative premise that the paper argues for but does not prove.
  • domain assumption Safety for AIG-NCII is determined by consent, not artificiality; authenticity tools therefore cannot be sufficient interventions
    Central conceptual claim built on Table 1 and qualitative arguments; plausible but not empirically demonstrated.
  • domain assumption Adversarial immunization and gated release can add sufficient friction to reduce abuse
    Recommendation-level assumption citing Jeon et al. 2025, Kim et al. 2026, and Van Le et al. 2023; the paper itself concedes brittleness in AV2.

pith-pipeline@v1.3.0-alltime-deepseek · 18663 in / 13267 out tokens · 127597 ms · 2026-08-02T13:46:10.651761+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Position: AI/ML Deepfake Research is Misaligned with AI-Generated Non-Consensual Intimate Imagery (AIG-NCII)." pith.science (2026). https://pith.science/paper/WNLEB4CW

@misc{pith2026260718263,
  author       = {Pith},
  title        = {Pith review of: Position: AI/ML Deepfake Research is Misaligned with AI-Generated Non-Consensual Intimate Imagery (AIG-NCII)},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WNLEB4CW}},
  note         = {Machine review of arXiv:2607.18263}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

AI-generated non-consensual intimate imagery (AIG-NCII) is not adequately addressed in AI/ML literature regarding AI-generated media, commonly referred to as "deepfakes". While research on deepfakes currently focuses on its epistemic harms -- or harms relating to truth and authenticity -- this is misaligned with the dominant reality of generative AI abuse involving sexualized imagery. We conduct a landscape analysis of highly-cited works to demonstrate that technical interventions addressing deepfakes almost entirely ignore AIG-NCII, limiting the research ecosystem to authenticity detection tools. In this position paper, we argue that existing interventions address viewer-centric epistemic harms, such as fraud or scams, but ignore subject-centric dignity harms, such as AIG-NCII. We illustrate that knowing an image is synthetic does not mitigate harms to subjects and may, in some cases, even exacerbate them. We conclude by offering recommendations to realign the field, including updating threat models to consider subject-centric harms and addressing AIG-NCII in AI safety research. Finally, we caution that researchers should only engage in this high-risk domain if they implement safety guardrails for both subjects and researchers and establish partnerships with domain experts in sexual violence prevention.

Figures

Figures reproduced from arXiv: 2607.18263 by Eric Gilbert, Li Qiwei, Sarita Schoenebeck, Wells Lucas Santo.

Figure 1
Figure 1. Figure 1 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Examples of deepfakes that cause harm to the viewer, subject, or both parties. 4. Consequences of ignoring AIG-NCII The intervention paradigms around detection, provenance, and watermarking are meant to restore the viewer’s ability to discern what is authentic. For AIG-NCII cases, however, the subject suffers from violation of their dignity regard￾less of authenticity. In this section, we discuss possible … view at source ↗

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

154 extracted references · 2 canonical work pages

  1. [1]

    2021 10th International Conference on Information and Automation for Sustainability (ICIAfS) , pages=

    Deepfakes detection methods: a literature survey , author=. 2021 10th International Conference on Information and Automation for Sustainability (ICIAfS) , pages=. 2021 , organization=

    2021

  2. [2]

    International Conference on Learning Representations , volume=

    Adversarial perturbations cannot reliably protect artists from generative ai , author=. International Conference on Learning Representations , volume=

  3. [3]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Erasing concepts from diffusion models , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  4. [4]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Safe latent diffusion: Mitigating inappropriate degeneration in diffusion models , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  5. [5]

    Ethics of data and analytics , pages=

    Machine bias , author=. Ethics of data and analytics , pages=. 2022 , publisher=

    2022

  6. [6]

    Conference on fairness, accountability and transparency , pages=

    Gender shades: Intersectional accuracy disparities in commercial gender classification , author=. Conference on fairness, accountability and transparency , pages=. 2018 , organization=

    2018

  7. [7]

    Stanford Internet Observatory, Cyber Policy Center, December , volume=

    Identifying and eliminating csam in generative ml training data and models , author=. Stanford Internet Observatory, Cyber Policy Center, December , volume=

  8. [8]

    The Fourteenth International Conference on Learning Representations , year=

    DiffVax: Optimization-Free Image Immunization Against Diffusion-Based Editing , author=. The Fourteenth International Conference on Learning Representations , year=

  9. [9]

    Advances in Neural Information Processing Systems , volume=

    BlurGuard: A Simple Approach for Robustifying Image Protection Against AI-Powered Editing , author=. Advances in Neural Information Processing Systems , volume=

  10. [10]

    a stalker's paradise

    “a stalker's paradise” how intimate partner abusers exploit technology , author=. Proceedings of the 2018 CHI conference on human factors in computing systems , pages=

    2018

  11. [11]

    arXiv preprint arXiv:2005.05535 , year=

    DeepFaceLab: Integrated, flexible and extensible face-swapping framework , author=. arXiv preprint arXiv:2005.05535 , year=

    Pith/arXiv arXiv 2005

  12. [12]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Adding conditional control to text-to-image diffusion models , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  13. [13]

    The Impact of AI-Generated Deepfake Social Media Content on Women in Bangladesh: A Qualitative Study , author=

  14. [14]

    2025 , note =

    Article 50: Transparency Obligations for Providers and Deployers of Certain AI Systems , author =. 2025 , note =

    2025

  15. [15]

    2024 , month =

    Our Approach to Labeling AI-Generated Content and Manipulated Media , author =. 2024 , month =

    2024

  16. [16]

    2026 , note =

    About AI-generated Content , author =. 2026 , note =

    2026

  17. [17]

    Cognition and Emotion , volume=

    Real is the new sexy: The influence of perceived realness on self-reported arousal to sexual visual stimuli , author=. Cognition and Emotion , volume=. 2024 , publisher=

    2024

  18. [18]

    Social & legal studies , volume=

    Sexual violence in the digital age: The scope and limits of criminal law , author=. Social & legal studies , volume=. 2016 , publisher=

    2016

  19. [19]

    Trauma, Violence, & Abuse , volume=

    Image-based sexual abuse perpetration: A scoping review , author=. Trauma, Violence, & Abuse , volume=. 2024 , publisher=

    2024

  20. [20]

    2025 , note =

    Faridabad Student Dies by Suicide After AI-Generated Deepfake Blackmail , author =. 2025 , note =

    2025

  21. [21]

    , author=

    Lora: Low-rank adaptation of large language models. , author=. ICLR , volume=

  22. [22]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  23. [23]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    High-resolution image synthesis with latent diffusion models , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  24. [24]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Conditional image-to-image translation , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  25. [25]

    Communications of the ACM , volume=

    Generative adversarial networks , author=. Communications of the ACM , volume=. 2020 , publisher=

    2020

  26. [26]

    Violation of my \ body: \

    " Violation of my \ body: \ " Perceptions of \ AI-generated \ non-consensual (intimate) imagery , author=. Twentieth Symposium on Usable Privacy and Security (SOUPS 2024) , pages=

    2024

  27. [27]

    arXiv preprint arXiv:2512.11815 , year=

    Video Deepfake Abuse: How Company Choices Predictably Shape Misuse Patterns , author=. arXiv preprint arXiv:2512.11815 , year=

    Pith/arXiv arXiv

  28. [28]

    Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

    A Closer Look at the Existing Risks of Generative AI: Mapping the Who, What, and How of Real-World Incidents , author=. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

  29. [29]

    proceedings of the 2022 international conference on multimodal interaction , pages=

    On the horizon: Interactive and compositional deepfakes , author=. proceedings of the 2022 international conference on multimodal interaction , pages=

    2022

  30. [30]

    arXiv preprint arXiv:2504.17663 , year=

    The Malicious Technical Ecosystem: Exposing Limitations in Technical Governance of AI-Generated Non-Consensual Intimate Images of Adults , author=. arXiv preprint arXiv:2504.17663 , year=

    arXiv

  31. [31]

    Information, Communication & Society , pages=

    ‘It's still abuse’: community attitudes and perceptions on AI-generated image-based sexual abuse , author=. Information, Communication & Society , pages=. 2026 , publisher=

    2026

  32. [32]

    Applications of Digital Image Processing XLV , volume=

    C2PA: the world’s first industry standard for content provenance (Conference Presentation) , author=. Applications of Digital Image Processing XLV , volume=. 2022 , organization=

    2022

  33. [33]

    arXiv preprint arXiv:2305.20030 , year=

    Tree-ring watermarks: Fingerprints for diffusion images that are invisible and robust , author=. arXiv preprint arXiv:2305.20030 , year=

    Pith/arXiv arXiv

  34. [34]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    The stable signature: Rooting watermarks in latent diffusion models , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  35. [35]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Towards universal fake image detectors that generalize across generative models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  36. [36]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Dire for diffusion-generated image detection , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  37. [37]

    International conference on machine learning , pages=

    Leveraging frequency analysis for deep fake image recognition , author=. International conference on machine learning , pages=. 2020 , organization=

    2020

  38. [38]

    Applied Sciences , volume=

    Deep fake image detection based on pairwise learning , author=. Applied Sciences , volume=. 2020 , publisher=

    2020

  39. [39]

    ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    On the detection of synthetic images generated by diffusion models , author=. ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2023 , organization=

    2023

  40. [40]

    First Monday , year=

    Nothing new here: Emphasizing the social and cultural context of deepfakes , author=. First Monday , year=

  41. [41]

    Deep fakes: A looming challenge for privacy, democracy, and national security , author=. Calif. L. Rev. , volume=. 2019 , publisher=

    2019

  42. [42]

    Yale LJ , volume=

    Sexual privacy , author=. Yale LJ , volume=. 2018 , publisher=

    2018

  43. [43]

    2022 , publisher=

    The fight for privacy: Protecting dignity, identity and love in the digital age , author=. 2022 , publisher=

    2022

  44. [44]

    Deepfakes and the epistemic backstop , author=

  45. [45]

    34th USENIX Security Symposium (USENIX Security 25) , pages=

    Analyzing the \ AI \ nudification application ecosystem , author=. 34th USENIX Security Symposium (USENIX Security 25) , pages=

  46. [46]

    2019 International Conference on Biometrics (ICB) , pages=

    Vulnerability assessment and detection of deepfake videos , author=. 2019 International Conference on Biometrics (ICB) , pages=. 2019 , organization=

    2019

  47. [47]

    34th USENIX Security Symposium (USENIX Security 25) , pages=

    Characterizing the \ MrDeepFakes \ Sexual Deepfake Marketplace , author=. 34th USENIX Security Symposium (USENIX Security 25) , pages=

  48. [48]

    arXiv preprint arXiv:2503.02250 , year=

    Ai automatons: Ai systems intended to imitate humans , author=. arXiv preprint arXiv:2503.02250 , year=

    Pith/arXiv arXiv

  49. [49]

    Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

    Stop the nonconsensual use of nude images in research , author=. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , volume=

  50. [50]

    2023 13th International Conference on Cloud Computing, Data Science & Engineering (Confluence) , pages=

    A comparative study: deepfake detection using deep-learning , author=. 2023 13th International Conference on Cloud Computing, Data Science & Engineering (Confluence) , pages=. 2023 , organization=

    2023

  51. [51]

    2020 25th international conference on pattern recognition (ICPR) , pages=

    Video face manipulation detection through ensemble of cnns , author=. 2020 25th international conference on pattern recognition (ICPR) , pages=. 2021 , organization=

    2020

  52. [52]

    Bond, Shannon , year = 2024, month = dec, journal =. How

    2024

  53. [53]

    , year = 2026, month = jan, journal =

    Conger, Kate and Freedman, Dylan and Thompson, Stuart A. , year = 2026, month = jan, journal =. Musk's

    2026

  54. [54]

    Deepfakes of

    New, Brian and Salazar, Lexi and Lozano, Mike and Fralicks, Scott , year = 2024, month = nov, journal =. Deepfakes of

    2024

  55. [55]

    arXiv preprint arXiv:2602.04759 , year=

    How to Stop Playing Whack-a-Mole: Mapping the Ecosystem of Technologies Facilitating AI-Generated Non-Consensual Intimate Images , author=. arXiv preprint arXiv:2602.04759 , year=

    Pith/arXiv arXiv

  56. [56]

    The Harm of Deepfakes: A Scoping Review of Deepfakes' Negative Effects on Human Mind and Behavior , shorttitle =

    Diel, Alexander and Lalgi, Tania and Mellis, Finley Sam and Teufel, Alexander and B. The Harm of Deepfakes: A Scoping Review of Deepfakes' Negative Effects on Human Mind and Behavior , shorttitle =. AI & SOCIETY , issn =. doi:10.1007/s00146-025-02774-0 , urldate =

    work page doi:10.1007/s00146-025-02774-0

  57. [57]

    404 Media , author =

    Hugging. 404 Media , author =

  58. [58]

    CivitAI , year =. Policy

  59. [59]

    2024 , author =

    South. 2024 , author =

    2024

  60. [60]

    Bicker, Laura , url =. Cho

  61. [61]

    arXiv preprint arXiv:2504.13959 , year=

    AI Safety Should Prioritize the Future of Work , author=. arXiv preprint arXiv:2504.13959 , year=

    Pith/arXiv arXiv

  62. [62]

    Gibson, Jenna , year =. Korea

  63. [63]

    Millions of

    Burgess, Matt , urldate =. Millions of

  64. [64]

    New media & society , volume=

    \# Gamergate and The Fappening: How Reddit’s algorithm, governance, and culture support toxic technocultures , author=. New media & society , volume=. 2017 , publisher=

    2017

  65. [65]

    Porn Studies , volume=

    ‘A Deepfake Porn Plot Intended to Silence Me’: exploring continuities between pornographic and ‘political’deep fakes , author=. Porn Studies , volume=. 2020 , publisher=

    2020

  66. [66]

    GOV.UK , file =

    Online. GOV.UK , file =

  67. [67]

    2025 , author =

    S.146 - 119th. 2025 , author =

    2025

  68. [68]

    Durbin, Richard J

    Sen. Durbin, Richard J. [D-IL , month = jul, year =. S.3696 - 118th

  69. [69]

    Deepfake ‘

    Burgess, Matt , file =. Deepfake ‘

  70. [70]

    Singer, Natasha , url =. Real

  71. [71]

    Available at SSRN 4641526 , year=

    Building the epistemic community of AI safety , author=. Available at SSRN 4641526 , year=

  72. [72]

    arXiv preprint arXiv:2512.13500 , year=

    Platforms as Crime Scene, Judge, and Jury: How Victim-Survivors of Non-Consensual Intimate Imagery Report Abuse Online , author=. arXiv preprint arXiv:2512.13500 , year=

    arXiv

  73. [73]

    Journal of Academic Ethics , volume=

    Secondary trauma: Emotional safety in sensitive research , author=. Journal of Academic Ethics , volume=. 2020 , publisher=

    2020

  74. [74]

    Nature Machine Intelligence , pages=

    AI safety for everyone , author=. Nature Machine Intelligence , pages=. 2025 , publisher=

    2025

  75. [75]

    Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems , pages=

    A Law of One's Own: The Inefficacy of the DMCA for Non-Consensual Intimate Media , author=. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems , pages=

    2025

  76. [76]

    arXiv preprint arXiv:2505.04600 , year=

    Perpetuating Misogyny with Generative AI: How Model Personalization Normalizes Gendered Harm , author=. arXiv preprint arXiv:2505.04600 , year=

    Pith/arXiv arXiv

  77. [77]

    We're utterly ill-prepared to deal with something like this

    " We're utterly ill-prepared to deal with something like this": Teachers' Perspectives on Student Generation of Synthetic Nonconsensual Explicit Imagery , author=. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems , pages=

    2025

  78. [78]

    404 Media , author =

    a16z-. 404 Media , author =. 2025 , file =

    2025

  79. [79]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Gaussian shading: Provable performance-lossless image watermarking for diffusion models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  80. [80]

    IEEE Access , volume=

    An improved dense CNN architecture for deepfake image detection , author=. IEEE Access , volume=. 2023 , publisher=

    2023

Showing first 80 references.