data-poisoning

This component is a high-risk, dual-use offensive artifact that implements concrete data-poisoning and backdoor techniques, including LLM poisoning, and provides operational advice to evade detection. Although the code is partly illustrative and contains some undefined variables, it is sufficiently detailed to be weaponized by anyone with ML expertise. Treat as potentially malicious-capability-bearing content: restrict usage to authorized, isolated testing environments, add governance controls (authorization, data provenance, integrity checks, auditing), and avoid deploying in production or distributing widely without safeguards.