A structured, evidence based map of how real AI systems get attacked, and what actually happens when they do.
Antijection documents and classifies real-world attacks on deployed AI systems, organized by where they happen, what enables them, and what they actually change.
What makes this different
Most frameworks classify attacks by attacker behaviour or generic risk category. None tell a team which attacks are actually possible in their specific deployment. Antijection links attack types to the specific external connections that enable them, so the threat profile follows the deployment.
The field measures attacks by success rate in lab benchmarks. That does not capture what changed in the real system. Antijection tracks observable outcomes: the system-state changes a security team would actually see in their logs.
Most documented attacks come from controlled research. Antijection documents what has genuinely happened in production, and how that diverges from theory.
Built as a living, structured, navigable knowledge base that grows as new incidents are documented and classified.
The map
incidents linked to the deployment it ran in, the connection it flowed through, the technique it used, and the outcome it produced. Switch between flow, matrix, timeline, and 3D scatter views below.
Want the full-screen view? /graph
Maintained by Aiteera LLC, an AI integration and development company.