mansaibots — agentic threat model

Not certain from the listing — likely utilizes third-party LLMs for customizable models. Primary threats include prompt injection to bypass system instructions and model alignment, potentially leading to brand damage or unauthorized instructions execution.

High risk. The platform ingests documents, websites, CSV/SQL data, and APIs for RAG. This exposes it to data poisoning (e.g., via scraped malicious websites) and indirect prompt injection embedded in databases or files, leading to unauthorized data access.

High risk. The agent orchestrates workflows and calls APIs to guide purchases and automate tasks. Insecure tool integration could allow an attacker to manipulate the agent via chat to trigger unauthorized API calls or database queries.

Not certain from the listing — assumed to be a multi-tenant SaaS deployment. Key threats include insecure storage of customer API keys/database credentials and lack of sandboxing for dynamic database queries.

Moderate risk. Provides chat history access for performance insights, but lacks explicit mention of real-time guardrails, anomaly detection, or automated prompt injection filtering.

Not certain from the listing — despite targeting sensitive industries like healthcare and e-commerce, there is no mention of compliance standards (e.g., HIPAA, GDPR, SOC2) or robust access control policies for database integrations.

Not certain from the listing — no explicit multi-agent orchestration or marketplace is described, though integration with external APIs and workflows introduces risks of cascading failures if downstream services are compromised.