What Is ZDX Text Guard? ZDX Text Guard is a deterministic text sanitation and detection system designed to protect AI and agent workflows before text ever reaches the model. Instead of relying on probabilistic moderation or hoping a model “notices” something is wrong, Text Guard enforces text integrity at the boundary—where security actually works. It runs locally, offline, and deterministically. No magic. No guessing. What Problems Does It Solve? ZDX Text Guard detects, sanitizes, or blocks entire classes of text manipulation techniques, including: Zero-width and invisible characters used to hide payloads Bidirectional override characters that visually reorder text Unicode tag characters capable of encoding invisible ASCII Combining character abuse (“zalgo” text) Unicode whitespace smuggling Structural impersonation such as fake <system> delimiters These techniques are increasingly used in prompt injection, agent manipulation, and tool abuse scenarios—especially in terminal-based AI workflows. How It Works ZDX Text Guard operates as a modular pipeline: Scan incoming text using deterministic rules Sanitize safe-but-obfuscated content Detect and block high-confidence manipulation Warn on suspicious but non-fatal patterns Pass only clean text to the AI system It is implemented today as a guard for Claude Code / terminal AI, with a simple on/off switch controlled entirely by the user. Why Deterministic Matters Security systems should be: Auditable Testable Predictable Text Guard ships with explicit acceptance tests covering real manipulation vectors, not vague heuristics. Every decision is explainable. Every block is intentional. This makes it suitable not just for individuals, but for teams, CI systems, and future API gateways. Part of a Bigger Vision ZDX Text Guard is not the end goal—it’s the foundation. At ZeroDriveX, we’re building toward a layered AI security model where: Deterministic guards eliminate known attack surfaces Lightweight AI classifiers handle semantic risk Agent behavior is constrained by enforceable contracts Tool execution is verifiable and auditable Text integrity is the first invariant. Read the White Paper We’ve published a full technical white paper detailing: The threat model The architecture The current implementation Test coverage And where ZDX is taking this next 👉 Read the ZDX Text Guard White Paper If you work with AI in terminals, agents, or production workflows, this is worth your time.

- ZerodriveX