Working map
The reviewed handoff research organizes Parasitic AI as a host-extraction lens, not as a settled scientific taxonomy. The strongest public version of the site should therefore stay bounded: identify the host, identify what the AI-mediated system gains, and measure whether the host or shared environment is degraded.
The five pathways
- Psychological and persona capture: sycophancy, emotional reliance, delusion reinforcement, and long-context escalation.
- Creative and cultural extraction: unlicensed training disputes, style imitation, attribution loss, and market substitution pressure.
- Search and retrieval pollution: scaled content abuse, site-reputation abuse, synthetic evidence, and retrieval-source homogenization.
- Data-ecosystem contamination: recursive synthetic-data loops, model collapse risk, and loss of tail information.
- Cyber and infrastructure exploitation: prompt injection, agent hijacking, AI-worm propagation, scams, malware assistance, and non-human identity misuse.
Boundary rule
Not every AI-generated artifact is parasitic. The line is crossed when the system gains persistence, distribution, or optimization advantage while the host loses autonomy, trust, source diversity, compensation, safety, or control.