About OpenAttacks
OpenAttacks is a community-maintained catalog of reported attacks and security research where AI is part of the attack path. Every entry keeps the evidence, uncertainty, and original researchers in view.
Built by defenders for defenders, the catalog turns public research into explorable attack flows, defensive context, draft hunts, and machine-readable data without replacing the original source.
Evidence before assumptions.
OpenAttacks helps researchers, incident responders, detection engineers, and security teams understand how AI is appearing in real attack paths. The project welcomes sourced contributions and makes the resulting catalog available through the website, public APIs, Markdown bundles, and an agent skill.
The catalog currently contains 275 entries spanning 2023–2026. It records documented coverage, not global attack prevalence or a risk ranking.
The people behind the project
Research, catalog stewardship, detection engineering, and open-source building.

Kostas
Kostas builds tools and research for threat hunting, threat intelligence, and detection engineering. His work includes Threat Hunting Labs, EDR Telemetry, and DetectionStream.

Michael Haag
Michael is MagicSword's CTO and co-founder. A former principal threat researcher at Splunk, he helped found Atomic Red Team, LOLDrivers, and LOLRMM.

Jose Hernandez
Jose is MagicSword's CEO and co-founder. He turns attacker techniques into practical defenses and contributes to projects including Atomic Red Team, LOLBAS, and Splunk Attack Range.

Nasreddine Bencherchali
Nasreddine focuses on detection engineering and threat hunting at Cisco/Splunk. He maintains Sigma and co-founded the EVTX-ETW-Resources project.

Scott Roberts
Scott is a network defender, developer, educator, and author of O'Reilly's Intelligence-Driven Incident Response. He brings more than 15 years of experience in threat research, detection, and incident response.
Read OpenAttacks with an AI assistant
The agent skill is the easiest entry point for Claude, Cursor, and other assistants. The same validated YAML also produces a concise index, per-attack Markdown, smaller category and month bundles, the full catalog, and structured JSON.
Contribute a documented attack
Attack records live in YAML in the public repository. Contributions can add sourced cases, correct metadata, improve defensive context, or link new research and detections. Uncertain details should say “unknown,” and submissions must not include live victim data or unredacted victim IP addresses.
Give your AI the evidence.
Complete source records, flows and draft hunts. Regenerated on every build.