AETHER Red Team Dataset
Uncensored offensive cybersecurity dataset for training autonomous pentesting AI (Hermes Agent).
v15 (2026-05-02)
- 20,072 redteam records (93.8% approved)
- Dropped: 2,687 non-redteam + 66 syntax errors + 4 refusals + 5 exact duplicates
- Categories: cloud, active_directory, ics_ot, malware_edr, web_api, cryptography, binary_exploitation, network_infra, osint_social_engineering, threat_intel_purple_team, mobile, bug_bounty, wireless_physical
- Prerequisites populated: 15,959 records
- MITRE ATT&CK phases: all 20,072 records
- Command syntax validated against real tool docs (PetitPotam, certipy, mimikatz, secretsdump, IMDSv2, AWS CLI, frida, sed, PowerView)
- Refusal content purged (ZERO refusals)
- Stratified splits: train 16,053 / val 2,002 / test 2,017
Splits
| Split | Records |
|---|---|
| Train | 16,053 |
| Val | 2,002 |
| Test | 2,017 |
Format
OpenAI-compatible messages with native tool_calls JSON + tool_call_id linkage.