Gemini 1.5
Google DeepMind — Gemini
rich detail 9 method fields published
Evaluation
Prompt injection evaluation (handcrafted and optimization-based attacks)
Vulnerability to prompt injection where attacker manipulates model to output a markdown image exfiltrating sensitive info from conversation history (six data categories); handcrafted templates vs genetic-algorithm optimization attacks
How it was run
Open a release to see the setup, scoring, and other details that the lab published. Only fields the lab actually disclosed are shown.