ARTCB TERMINATOR is a KYC Clone — a KYC AI agent designed to reproduce the workflow of a bank’s KYC officer.
KYC processes are repetitive, time-consuming, and highly sensitive. Automating them with AI can improve efficiency, but it also introduces new risks: agents may rely on incomplete information, lose workflow context, or be manipulated by malicious instructions hidden in documents or other inputs.
Our KYC Clone addresses these challenges through three core capabilities: Remember, Know, and Prove.
Remember: Powered by Screenpipe, the Clone recovers workflow context from screen activity, allowing it to understand how a KYC process is performed and maintain continuity across tasks.
Know: The Clone completes KYC questionnaires using only available company data and evidence. When information is missing, it does not invent an answer, helping keep outputs reliable and reviewable.
Prove: Guardian protects the agent from prompt-injection attacks by intercepting and blocking unsafe actions before execution. The process is then recorded in a verifiable audit trail, with SHA-256 chaining used to preserve evidence of what happened.
ARTCB TERMINATOR is designed to reduce repetitive KYC work while keeping AI actions controlled, traceable, and auditable. The bank remains in control of the final decision, while the KYC Clone assists with the operational work needed to reach it.
ARTCB TERMINATOR is built around a Python and Rust architecture. Python handles AI-agent orchestration, model interactions, tool integrations, the Guardian API, and controlled attack scenarios. Rust provides the security core for policy enforcement, event integrity, provenance tracking, evidence verification, and replay.
Screenpipe provides the workflow-memory layer. It captures observable desktop activity and gives the KYC AI agent contextual information about the operator’s workflow without treating that activity as authoritative business evidence.
A dedicated /api/screenpipe/context endpoint retrieves relevant Screenpipe events and normalizes fields such as timestamp, application, text, and event type. When Screenpipe is unavailable, an explicitly identified offline fallback allows the demonstration to continue without presenting simulated activity as live data.
The architecture separates three types of information: observed context from Screenpipe, trusted facts used by the KYC agent, and security evidence produced by Guardian.
Screenpipe provides memory, but never authorization. Proposed sensitive actions pass through Guardian before execution, where policies can allow or block them. Security events are then recorded for integrity, provenance, and replay.
In short: REMEMBER with Screenpipe, KNOW with trusted data, and PROVE with Guardian.