Dream Companion — agentic threat model

Uses LLMs and image generation models (e.g., Stable Diffusion) optimized for unfiltered NSFW content. Primary threats include model reprogramming to generate illegal/harmful content and model exploitation via adversarial prompts.

Not certain from the listing — No details on vector databases or RAG are provided, but the system must store user chat history and custom character profiles. The primary threat is the exfiltration of highly sensitive, personally identifiable NSFW chat logs and custom character data.

Not certain from the listing — The orchestration framework is unspecified but likely manages character state, memory persistence, and routing to the image generator. Threats include memory poisoning and insecure state management leading to cross-user data leakage.

Not certain from the listing — As an open-source project, deployment is user-dependent, but if hosted as a service, threats include container compromise, unauthorized access to GPU resources, and exposure of image generation APIs.

Not certain from the listing — The 'unfiltered' and 'uncensored' nature suggests a deliberate lack of safety guardrails, creating a significant blind spot for monitoring toxic, non-consensual, or illegal content generation.

Not certain from the listing — No compliance certifications or strict identity/access management are detailed. The agent faces high compliance risks regarding data privacy regulations (GDPR/CCPA) due to the highly sensitive nature of user-generated adult content.

Operates as a standalone vertical application with no multi-agent or marketplace interactions described. Threats in this layer are minimal, limited to the manual sharing of custom character templates.