Skip to content

Training Data Poisoning

Named by 3 kept items in this workspace.

Search every workspace for Training Data Poisoning. It matches the name as written, so it can find other things that share it.

The items that name it 3

  • elvis (@omarsar0) on X
    thread2026-10-30

    Poisoning 0.1% of pretraining data planted a remote code execution backdoor in a CLI agent. At a 0.0005% poison rate (about 2,500 documents in a 100B-token corpus), the trigger still fired on 1.4% of prompts in a single-seed run.

  • requie/AI-Red-Teaming-Guide
    repo2026-07-08

    Injecting malicious/poisoned examples into training data; backdoor, availability, targeted, and clean-label poisoning attacks

  • MITRE ATLAS v2026.09 data release
    repo

    Previously called 'Poison Training Data', v2026.07

All techniques in this workspace.