← Search

Ethan Kosak-Hine

2 accepted papers

2025

SandboxSocial: A Sandbox for Social Media Using Multimodal AI Agents

IJCAI 2025

The online information ecosystem enables influence campaigns of unprecedented scale and impact. We urgently need empirically grounded approaches to counter the growing threat of malicious campaigns, now amplified by generative AI. But, developing defenses in real-world settings is impractical. Socia

2025

The Structural Safety Generalization Problem

ACL 2025finding

LLM jailbreaks are a widespread safety challenge. Given this problem has not yet been tractable, we suggest targeting a key failure mechanism: the failure of safety to generalize across semantically equivalent inputs. We further focus the target by requiring desirable tractability properties of atta…