Economic Espionage

AI assets represent extraordinary value concentration: a frontier model represents $100M+ in training compute, proprietary training datasets may be irreplaceable, and AI-derived business intelligence (pricing algorithms, recommendation logic, customer models) is competitively decisive. Economic espionage targets this value through model extraction (query-based distillation), training data extraction (membership inference, data reconstruction), trade secret exfiltration through AI-assisted intelligence gathering, and insider threat amplified by AI tools. The trust assumption violated is that serving a model through an API doesn't leak the model itself — in practice, systematic querying can extract a functional copy (model stealing) or reconstruct training data (training data extraction).

Economic espionage chains from T14-AT-012 (Cloud Provider Exploitation) for infrastructure access and T14-AT-001 (GPU Farm Hijacking) for direct model weight access. Chains into T14-AT-006 (Competitive Sabotage) when stolen intelligence is used to undercut the victim.