TotalAgility — agentic threat model

Not certain from the listing — The specific foundation models (LLMs) powering the copilot and document processing are not disclosed. Threats include model-level evasion, prompt injection bypassing low-code boundaries, and potential data leakage if public APIs are used without enterprise agreements.

Highly critical layer. TotalAgility builds advanced RAG patterns, hosts knowledge bases, and processes diverse document types. This exposes the platform to document-based prompt injection, knowledge base poisoning, and unauthorized data exfiltration via RAG retrieval mechanisms.

The platform orchestrates agents, RPA, and workflows using natural language. Insecure tool integration is a major threat here, as malicious instructions embedded in processed documents could trigger unauthorized RPA actions or API calls within the process orchestration engine.

Not certain from the listing — The hosting architecture (cloud vs. on-premise), containerization, and sandboxing of RPA execution environments are not detailed. Inadequate sandboxing could allow an agent executing a malicious script to compromise the host environment.

Strongly addressed via 'Agent Governance, Testing, Benchmarking & Audit Trails' and source document highlighting. However, risks remain regarding the detection of subtle semantic drift in document extraction models or adversarial manipulation of the benchmarking suite.

Features built-in governance, audit trails, and 'Human in the Loop' case management. The primary threat is the bypass of these controls if authorization policies between the low-code agent builder and the underlying RPA execution roles are misconfigured.

As an agent-building platform, it facilitates the creation of multiple interacting automations. Threats include cascading failures across automated workflows and unauthorized horizontal escalation if one agent's output is implicitly trusted by another downstream process.