Add dataset card
#2
by nielsr HF Staff - opened
README.md
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
task_categories:
|
| 3 |
+
- text-classification
|
| 4 |
+
tags:
|
| 5 |
+
- security
|
| 6 |
+
- llm-agents
|
| 7 |
+
- prompt-injection
|
| 8 |
+
- safety
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# ShadowMem Dataset
|
| 12 |
+
|
| 13 |
+
This dataset is part of the artifacts released for the paper [Safeguarding LLM Agents against Long-Horizon Threats via Shadow Memory](https://huggingface.co/papers/2605.03228).
|
| 14 |
+
|
| 15 |
+
**ShadowMem** is a defensive framework that maintains a dedicated, safety-focused agentic memory — inspired by the *shadow stack* abstraction in systems security — to distill and retain safety-critical context across an agent's full execution trajectory, and uses this shadow memory to proactively assess pending actions before they are executed.
|
| 16 |
+
|
| 17 |
+
Code: https://github.com/ZJUWYH/ShadowMem
|