Prompt-injection defense for documents

Stop hidden prompt injection
before it reaches your AI.

PromptRisk scans PDF, Word and Excel files for the tricks attackers hide inside documents — invisible text, smuggled Unicode, instruction overrides — and scores each file from 0 to 100.

Three ways to scan

The same detection engine, wherever your documents live.

🔍

Web app

Drag and drop a file and get an instant risk score, classification and the exact evidence found.

🔑

Paid API

Create an API key and call /api/v1/analyze from your own pipeline. Every call is metered.

☁️

Cloud folder scan

Connect Google Drive, Dropbox or OneDrive and scan an entire folder in one batch job.

A risk score you can act on

Every document is scored 0–100 and classified into five bands.

0–20SAFE
21–40LOW
41–60MODERATE
61–80HIGH
81–100CRITICAL

What PromptRisk looks for

A hybrid engine — deterministic checks plus a semantic LLM pass — covering 12 attack patterns.

Typographic camouflage (invisible text)

Metadata manipulation

Explicit instruction override

Typoglycemia / linguistic obfuscation

Unicode smuggling

Image steganography

Prompt leakage

Role-play / persona switching

Multi-turn manipulation

Delimiter confusion

Encoding / technical obfuscation

Data exfiltration / external callback

Scan your first document free

Create an account and analyze files in seconds. No credit card required.

Get started