SAI LabsBook a demo

Safety IndexEvaluationsCyberSecEval 4 multilingual injection

CyberSecEval 4 multilingual injection

Adversarial robustness · Prompt injectionCounts in the indexBuilt, not yet run

Whether the model resists injected instructions written in many languages.

Metric
% resisted. Higher is better.
Dataset
github.com/meta-llama/PurpleLlama · 1,004 · MIT (repo)
Grader
LLM judge (GPT-4.1) (the paper used same)
Status
Built, not yet run
Replicability
public

Results

Built and tested; the first full runs are pending.