01 // FILESDownload
Rebuilt with every update
Last updated 2026-11-17
Full dataset
Every harness with its use tags and the exact sentence behind each, every primary report, the tier definitions and the dates. The site renders from the same data.
Harnesses
One row per harness for a spreadsheet: tier, repo, GitHub facts, the tags seen in attacks and the tags from each README, actors and source ids.
Primary reports
Every report the index cites, with publisher, kind, date, title and link.
02 // FIELD GUIDEWhat each harness record holds
One record per harness
in data.json and harnesses.csv
| Field | Type | Meaning |
|---|---|---|
name | text | The harness name as the project or the report gives it. |
table | evidence | field | evidence: seen in a primary attack report. field: no attack on record. |
tier | text | confirmed, observed, or no attack on record. The tiers are defined in the file itself. |
repo | url or null | The GitHub repository. Null when the reports name a project with no canonical public repo. |
description / what_it_did | text | What the harness is, and for attack rows what the reports say it did. |
stars, license, last_push, archived | GitHub facts | Pulled from the GitHub API on the date in github_facts. |
uses[] | list | Use tags. Each has the tag, its basis (attack report or project docs), the https source, and the exact sentence behind it. |
actors[] | list | Attack rows only: the groups or operators the reports name, with the report's own description. |
breach, key_figure | text | Attack rows only: whether a report ties a breach to the harness, and the headline number. |
url | url | The harness row on the index, offsecharnesses.ai/#h-<name>, for linking one row. |
incidents[] | list | Incidents on hackingincidents.ai that name this harness: id, title, date and link, as of incidents_snapshot. |
sources[] | ids | Attack rows only: ids into the top-level sources list (publisher, date, title, url). |
03 // USE TAGSThe fixed set
Harness counts per tag, by where
the tag comes from
| Tag | Seen in attacks | From README | Meaning |
|---|---|---|---|
recon Recon | 6 | 70 | Maps a target: subdomains, ports, services, assets, OSINT. |
scan Scanning | 6 | 50 | Runs existing scanners and security tools (nmap, nuclei, sqlmap and the like) and reads their output. |
vuln Vuln discovery | 4 | 70 | Finds vulnerabilities by its own analysis: code review, fuzzing, logic, beyond a signature scanner. |
exploit Exploitation | 9 | 50 | Attacks a live target to prove or use a weakness. |
exploitdev Exploit dev | 0 | 36 | Writes exploits, proof-of-concept code or attack tooling. |
postex Post-exploitation | 7 | 35 | Works after access: credentials, privilege escalation, lateral movement, persistence, exfiltration. |
research Research | 0 | 20 | Academic paper code, a benchmark, or a CTF research harness. |
04 // MIRRORSubstrate
The same dataset
with the research notes
Data/Offensive-AI-Harnesses
The dataset mirrored in Substrate as JSON and CSV, with a summary, an update log and the script that refreshes it from this site.
Research notes
The research behind the rows, including what the page leaves out: commercial coding agents used as harnesses, dark-LLM services, and the record of the discovery sweep.
AI Hacking Incidents
The attack ledger this index links to. Each harness record here carries the incidents there that name it, under incidents[], as of incidents_snapshot.
How it is built
Where each kind of fact comes from, how a harness is tied to an attack, and the references used to find candidates.
Offsec Harness Index, offsecharnesses.ai, dataset updated 2026-11-17.