datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
b3-agent-security-benchmark-weak[paper] [blogpost] [game]
b3 AI Security Benchmark: Breaking Agent Backbones
Highly contextalized prompt injections crowd-sourced during the Gandalf Agent Breaker Challenge.
This is a low-quality version of the data behind Breaking Agent Backbones: Evaluating the Security
of Backbone LLMs in AI Agents.
The high quality dataset was used to evaluate the security of more than 30 LLMs.
Dataset Summary
Purpose: This dataset contains crowdsourced adversarial attacks… See the full description on the dataset page: https://huggingface.co/datasets/Lakera/b3-agent-security-benchmark-weak.ai-agent-security-incidents
AI Agent Security Incident Database v0.1
A structured, machine-readable database of 1365 confirmed AI agent security incidents, collected and classified automatically.
What is this?
Every time an AI agent causes unintended harm — escaping a sandbox, exploiting an API, taking unauthorized actions, exfiltrating data — this database captures it.
This is not a list of theoretical risks. Every entry describes something that actually happened, with a verifiable source… See the full description on the dataset page: https://huggingface.co/datasets/gemmozero/ai-agent-security-incidents.agent-skills-security-grades
Agent Skills Security Grades
Security grades and quality scores for 130,173 open-source AI agent skills and
MCP servers collected from GitHub, from Agent Skills Hub.
Each row is one skill/server with a rule-based security grade
(SAFE / CAUTION / UNSAFE / REJECT / UNAUDITED), red-flag identifiers, and a
0–100 quality score.
Why this exists
AI coding agents install third-party skills that run with the agent's full
permissions and credentials, but marketplaces rank… See the full description on the dataset page: https://huggingface.co/datasets/jasonzhuyansen/agent-skills-security-grades.ai-agent-security-sft-dpo
AI Agent Security — SFT + DPO
Fine-tuning data for teaching an AI agent to protect its confidential configuration without
becoming uselessly over-cautious. Built for
thesreedath/gemma-2-2b-qa-sft and
derived from
Dhanjo/ai-agent-security-dataset.
Why the helpfulness axis exists
leakage_score in the source dataset is one-sided: a model that refuses every request
scores a perfect 0.0. An existing fine-tune reported 0.0114 mean leakage (down from 0.4611
baseline)… See the full description on the dataset page: https://huggingface.co/datasets/sumitguha13/ai-agent-security-sft-dpo.ai-agent-security-dataset
AI Agent Security and System Prompt Leakage Dataset
Dataset Overview
This dataset was created for research on AI agent security, with a specific focus on system prompt leakage, jailbreak resistance, and security-aligned fine-tuning.
The dataset evaluates how often AI agents reveal confidential information embedded inside their system prompts when exposed to adversarial prompts. It also compares the behavior of a baseline language model against a model fine-tuned using… See the full description on the dataset page: https://huggingface.co/datasets/Dhanjo/ai-agent-security-dataset.