Agent Smith

Evals

Benchmark / dataset

One image can jailbreak many multimodal agents (infectious jailbreak).

Category
Prompt injection & agent security
Insurance exposure
Correlated/systemic
Publisher
Sea AI Lab
Publisher type
Academic
License
MIT
Status
Stale (>12mo)
Last commit
2024-03-26
Stars
130
Notes
Relevant to correlated loss.

View source on GitHub →

Wanich Risk Lab, 2026.