PromethAI — agentic threat model

Not certain from the listing — likely relies on commercial LLM APIs or local open-source models. The primary threat is prompt injection bypassing safety guardrails, potentially leading to the generation of harmful or allergen-triggering dietary recommendations.

Not certain from the listing — requires a database or vector store to track user goals, weight, and dietary preferences. Threats include unauthorized access or exfiltration of this sensitive personal health data (PHD) if stored insecurely.

Not certain from the listing — likely uses standard orchestration frameworks to parse user goals into actionable plans. Threats include insecure tool integration if the agent fetches external recipes or nutrition data from untrusted web sources.

Not certain from the listing — as an open-source tool, deployment is self-hosted or local. Threats include weak local host security, exposed API keys in environment files, and lack of network isolation for the running container.

Not certain from the listing — open-source personal tools rarely ship with built-in enterprise observability. The main threat is a lack of audit logs to detect if the assistant has been compromised or is consistently outputting hallucinated health advice.

Not certain from the listing — compliance with health data standards (like HIPAA) or privacy regulations (GDPR) is entirely dependent on the user's deployment environment, as the software likely lacks native multi-tenant access controls.

Not certain from the listing — operates primarily as a single-user standalone assistant. Ecosystem threats are minimal unless it is configured to interact with external third-party APIs or smart-home integrations.