Research Project

The advancement and integration of technology has raised concerns about technology-facilitated gender-based violence (TF-GBV). The incorporation of artificial intelligence (AI), specifically language learning models (LLMs) in daily life, could amplify gender-based violence and discrimination through the use of these technologies. This study explores how private conversations reflect harmful attitudes towards women. It also examines the degree to which AI could be used to promote gender-based violence and discrimination through the act of jailbreaking, or successfully using LLMs to generate material against its restrictions. Inductive content analysis was applied on 120 conversations derived from the WildChat database. The research found that a significant number of conversations contained language involving stereotyping, sexualization, objectification, and physical violence. Whereas companies emphasize systems with built-in safeguards to ensure safe outputs, the findings suggest that some constraints can be strategically negotiated or “jailbroken” in approximately 52.5% of the sample cases. These results indicate that LLMs can promote existing gender-based harms.