Looks like an ordinary document or webpage
Clean layouts and normal copy can still hide instructions in comments, hidden styles, metadata fields, or off-page content.
Human view versus model view
The risky text usually is not visible.
A page or file can look harmless to a person and still contain parser-visible or metadata-level instructions. Veridicus Scan exists for that gap between what a human sees and what a parser, document extractor, or model-ready pipeline actually receives.
The core gap
What looks ordinary to a person can still be risky to a model.
The problem is not abstract. Hidden DOM blocks, parser-only strings, suspicious metadata, hidden Unicode controls, and redirect behavior all change what downstream AI systems may actually consume.
Parsers ingest more than the eye sees
Parser-visible text, hidden Unicode, PDF or DOCX style signals, and URL redirects can all change what downstream AI receives.
You need evidence, not instinct
Risk has to end in a reportable answer: what was found, where it appeared, and whether the content should move forward.
How the gap shows up
Visibility, parsing, and decisions are three different layers.
The app is built to keep those layers explicit. That prevents the common failure mode where a person trusts a clean-looking page, while the actual extraction path feeds something very different into the next AI step.
Visible layer
What the person sees may be incomplete when content hides instructions in non-visible markup, styles, metadata, or layout tricks.
Normalized layer
Parsers, document readers, and URL fetch paths may expose hidden channels that are not obvious in the original visual presentation.
Decision layer
The safe choice depends on a readable report with findings, guidance, and coverage notes instead of a vague judgment.
Where this matters most
The gap shows up across web, documents, and agent workflows.
The same hidden-channel problem appears in public websites, imported files, and automated local agent flows. The container changes, but the risk pattern stays the same.
Assistant safety
Inspect a page or file before it becomes prompt context, retrieval material, or an uploaded attachment in an AI workflow.
Exact coverage
See the specific source types, hidden channels, and report outputs the app actually covers today.
MCP workflows
Use the same scanning and guardrail logic inside local, session-based MCP flows while the app is active.
Next step
Read what the app actually covers after reading why the problem matters.
Use the coverage page for the concrete scan surfaces, hidden channels, and decision outputs behind this problem statement.