A living research observatory for host-extractive intelligence
A cinematic archive mapping how AI-mediated systems can reproduce, entrench, mutate, or spread by extracting from hosts: attention, trust, creative labor, ranking signals, training data, infrastructure, and defenses.
Definition
Parasitic AI is framed here as an AI system, AI-mediated interaction, or AI-saturated environment that sustains itself by extracting from a host while degrading that host’s autonomy, epistemic environment, market position, or security posture.
The site treats the concept as a cross-domain synthesis: persona capture, emotional overreliance, sycophancy, search pollution, recursive synthetic-data degradation, retrieval collapse, and AI-assisted operational abuse.
Interactive demo
Click each stage to watch the conceptual vector shift from behavioral entry point to environmental propagation.
01 / 05
A prompt, interface pattern, companion persona, content farm, or agent instruction attaches to an existing host workflow without initially appearing hostile.
Taxonomy
Sycophancy, emotional reliance, delusion reinforcement, and relational loops that can narrow a user’s real-world options.
Models and markets that consume creator labor, style, and corpora while weakening attribution, revenue, and bargaining power.
Scaled synthetic pages parasitize ranking signals, then contaminate retrieval layers with low-value or adversarial evidence.
Recursive synthetic data loops degrade source diversity, erase rare information, and drive model-collapse dynamics.
Agents, worms, prompt injection, scams, phishing, malware generation, and lateral movement exploit trust and automation.
Containment lab
Adjust host dependency, propagation pressure, synthetic contamination, and containment strength to see a conceptual risk score. This is an educational visualization, not a production security model.
Defense doctrine
Evaluate whether conversational history creates dependency, delusion reinforcement, or unsafe escalation across many turns.
Track synthetic-to-real ratios, retrieval-domain entropy, tail-topic retention, and repeated style concentration.
Treat agents as privileged identities. Enforce least privilege, behavioral drift alerts, and microsegmented RAG stores.
Scan retrieved content and generated output for self-replication, instruction leakage, prompt injection, and data exfiltration cues.
Research archive
Install and activate the included Parasitic AI Core plugin to seed custom research-node posts and interactive shortcodes.