---
license: cc-by-nc-4.0
doi: 10.5281/zenodo.18157450
extra_gated_prompt: >-
  By requesting access to this dataset, you agree to:
  
  1. Use this dataset only for research, journalism, policy, or education purposes
  2. NOT attempt to re-identify hashed usernames
  3. NOT use this data to train AI systems to generate non-consensual imagery
  4. Cite this dataset in any publications
  5. Report any ethical concerns to [nana.nwachukwu@gmail.com]
extra_gated_fields:
  Name: text
  Affiliation: text
  Intended use: text
  I agree to the terms above: checkbox
task_categories:
  - text-classification
language:
  - en
tags:
  - TfGBV
  - AI safety
  - content moderation
  - NCII
  - gender-based violence
  - Grok
  - deepfakes
pretty_name: TfGBV-Grok Dataset
size_categories:
  - n<1K
---

# TfGBV-Grok Dataset: AI-Facilitated Non-Consensual Intimate Image Generation

> ⚠️ **WORK IN PROGRESS**: This dataset is actively being developed. There may be classification errors or inconsistencies. It will be updated as I process the rest of the raw data. Feedback and corrections are welcome — please email nana.nwachukwu@gmail.com.

## Dataset Description

This dataset documents **565 instances** of users publicly requesting Grok AI (integrated into X/Twitter) to generate non-consensual intimate imagery — digitally "undressing" women without their knowledge or consent.

The dataset was curated to support research on **technology-facilitated gender-based violence (TfGBV)** and to inform policy responses to emerging AI harms.

## Dataset Summary

| Statistic | Value |
|-----------|-------|
| Total records | 565 |
| Perpetrator requests | 481 |
| Prompt sharing (attack methods) | 48 |
| Victim testimonies | 7 |
| Reports of abuse | 10 |
| Commentary | 13 |
| Benign/unrelated Grok use | 6 |

## Actor Type Classification

Each record is classified by **who** is speaking:

| Actor_Type | Count | Description |
|------------|-------|-------------|
| `PERPETRATOR_REQUEST` | 481 | Users directly requesting Grok to generate NCII |
| `PROMPT_SHARING` | 48 | Users sharing attack methods/prompts (JSON exploits, etc.) |
| `COMMENTARY` | 13 | Users discussing or criticizing the phenomenon |
| `REPORTING_ABUSE` | 10 | Users calling out or questioning harmful behavior |
| `VICTIM_TESTIMONY` | 7 | Victims reporting that their images were targeted |
| `BENIGN_GROK_USE` | 6 | Unrelated Grok use (e.g., Halloween costumes) |

## Harm Type Classification

Each record is also classified by the **type of harm**:

| Harm_Type | Count | Description |
|-----------|-------|-------------|
| `NCSII_REQUEST` | 418 | Direct requests to remove/replace clothing |
| `SHARES_NCSII_METHOD` | 42 | Sharing prompts/methods for NCII generation |
| `DISCUSSES_NCSII` | 28 | Discussion of the NCII phenomenon |
| `PROMPT_INJECTION` | 8 | Structured JSON attacks to bypass safeguards |
| `NOT_NCSII_RELATED` | 6 | Benign content, not related to NCII |
| `SHARES_BODY_MOD_METHOD` | 6 | Sharing methods for body modification requests |
| `BODY_MODIFICATION` | 5 | Requests to alter body features |
| `DISCUSSES_MINOR_TARGETING` | 2 | Discussion of minors being targeted |

## Data Fields

| Column | Type | Description |
|--------|------|-------------|
| `Record_ID` | int | Unique identifier (1-565) |
| `Tweet_ID` | int | Original tweet ID — retained for verification |
| `Username_Hash` | string | SHA-256 hash of username (truncated to 12 characters) |
| `Date` | string | Standardized date (YYYY-MM-DD) |
| `Content_Snippet` | string | Text content of the tweet |
| `Actor_Type` | string | Who is speaking (perpetrator, victim, commentator, etc.) |
| `Harm_Type` | string | Type of harm documented |

Important Note on Classification
This dataset distinguishes between:

- Perpetrators — users actively requesting or sharing methods for NCII generation
- Victims — users reporting that theirs/ someone else's images were targeted
- Commentators — users discussing, criticizing, or reporting the phenomenon

This distinction is crucial for accurate research. Not every record represents a perpetrator request — some document the broader ecosystem of harm, including victim experiences and community responses.

## Intended Uses

✅ **Permitted uses:**
- Academic research on TfGBV and AI harms
- Policy analysis and advocacy
- Journalism and public interest reporting
- AI safety tool development (detection, moderation)
- Educational purposes

❌ **Prohibited uses:**
- Re-identification of hashed usernames
- Harassment or doxxing of any individuals
- Training AI systems to generate NCII
- Commercial use without authorization
- Redistribution without permission

## Access

This dataset is **gated** to ensure responsible use. By requesting access, you agree to:

1. Use the data only for permitted purposes listed above
2. NOT attempt to re-identify individuals
3. NOT use the data to develop tools that facilitate image-based abuse
4. Cite this dataset in any publications or outputs
5. Report any ethical concerns or incidents

## Ethical Considerations

### Privacy
- Usernames have been pseudonymized via SHA-256 hashing
- Tweet IDs are retained for verification purposes only
- No victim names or identifiers were found in the dataset

### Sensitivity
- This dataset documents harmful behavior and contains references to sexual content
- Two records reference minors — handle with appropriate care
- Content may be distressing to some researchers

### Limitations
- Dataset captures only publicly visible posts
- May not represent the full scope of the phenomenon
- Content snippets are partial, not full tweets

## Citation

```bibtex
@dataset{tfgbv_grok_2026,
  author = {[Nana Mgbechikwere Nwachukwu]},
  title = {TfGBV-Grok Dataset: AI-Facilitated Non-Consensual Intimate Image Generation},
  year = {2026},
  doi: 10.5281/zenodo.18157450
  publisher = {Zenodo},
  url = {https://huggingface.co/datasets/[Mtechlaw]/TfGBV-Grok-NCII-Dataset}
}
```

## Contact

For questions, access requests, or to report concerns:

**[Nana Mgbechikwere Nwachukwu]**  
Email: [nana.nwachukwu@gmail.com]

## Changelog

- **v2.0** (January 2026): Added 'Actor_Type' classification to distinguish perpetrators from victims, commentators, and reporters. Improved 'Harm_Type' taxonomy.
- **v1.0** (January 2026): Initial release