Competitive Sabotage

AI competitive advantage is fragile — model quality depends on training data, serving infrastructure, and user trust, all of which can be degraded by a competitor with offensive capability. Competitive sabotage targets these dependencies: poisoning public training datasets that competitors scrape, extracting proprietary models through query-based distillation, injecting backdoors into shared ML libraries, or attacking recommendation systems to degrade output quality. The trust assumption violated is inter-organizational: companies trust shared data sources, open-source libraries, and public APIs that competitors can manipulate.

Competitive sabotage chains from T6 (Training & Feedback Poisoning) for data poisoning techniques and T10 (Integrity & Confidentiality Breach) for model extraction. Chains into T14-AT-013 (Economic Espionage) when extracted models or intelligence provide competitive advantage.