kill-chain
kill-chain-canaries
Kill-Chain Canaries — run logs
Per-run outcomes from the paper
Kill-Chain Canaries: Stage-Level Tracking of Prompt Injection Across Attack
Surfaces and Model Safety Tiers
(arXiv:2603.28013).
Code: https://github.com/KevinChunye/prompt_injection
Each row is one benchmark run. A unique canary token is planted at an attack
surface, and the run records how far that token travelled through a two-agent
system — rather than only whether the attack ultimately "worked". That
stage-level… See the full description on the dataset page: https://huggingface.co/datasets/kevinwhc/kill-chain-canaries.kill-chain-forensics-training-data