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. This dataset documents 565 records related to the use of Grok AI (integrated into X/Twitter) for non-consensual intimate image generation — digitally "undressing" women without their knowledge or consent. The dataset includes perpetrator requests, victim testimonies, commentary, and reports of abuse — each clearly labeled to enable nuanced research. KEY STATISTICS:- 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:- PERPETRATOR_REQUEST — Users directly requesting Grok to generate NCII- PROMPT_SHARING — Users sharing attack methods/prompts (JSON exploits, etc.)- COMMENTARY — Users discussing or criticising the phenomenon- REPORTING_ABUSE — Users calling out or questioning harmful behaviour- VICTIM_TESTIMONY — Victims reporting that their images were targeted- BENIGN_GROK_USE — Unrelated Grok use (e.g., Halloween costumes) HARM TYPE CLASSIFICATION:- NCSII_REQUEST — Direct requests to remove/replace clothing- SHARES_NCSII_METHOD — Sharing prompts/methods for NCII generation- DISCUSSES_NCSII — Discussion of the NCII phenomenon- PROMPT_INJECTION — Structured JSON attacks to bypass safeguards- BODY_MODIFICATION — Requests to alter body features- DISCUSSES_MINOR_TARGETING — Discussion of minors being targeted DATASET FIELDS:- Record_ID: Unique identifier- Tweet_ID: Original tweet ID (for verification)- Username_Hash: SHA-256 hash of username (pseudonymized)- Date: Standardised date (YYYY-MM-DD)- Content_Snippet: Text of the tweet- Actor_Type: Who is speaking (perpetrator, victim, commentator, etc.)- Harm_Type: Type of harm documented ACCESS:Due to the sensitive nature of this data, the full dataset is available via gated access on HuggingFace:https://huggingface.co/datasets/Mtechlaw/TfGBV-Grok-NCII-Dataset Researchers must agree to a Data Use Agreement before access is granted. PERMITTED USES:- Academic research on TfGBV and AI harms- Policy analysis and advocacy- Journalism and public interest reporting- AI safety tool development- Educational purposes PROHIBITED USES:- Re-identification of pseudonymized usernames- Harassment or doxxing- Training AI to generate NCII- Commercial use without authorisation CONTACT:Nana Mgbechikwere [e-mail]