Android · by AppLabs
Check a file for hidden AI instructions before you share it.
PromptShield looks inside documents and images for prompt injection — text written to manipulate an AI tool that later reads the file. It scans on your phone, with no internet access, and shows you the evidence.
Coming soon to Google Play
What it looks for
Hidden text
White-on-white text, tiny fonts, hidden formatting, off-slide shapes, zero-width and invisible Unicode characters.
Places people don't look
Comments, notes, headers and footers, tracked deletions, alt text, document properties and image metadata.
Text in images
On-device text recognition reads screenshots, photos and scanned PDF pages.
Known attack patterns
Instructions to ignore previous rules, role and authority spoofing, attempts to exfiltrate data, and encoded payloads.
Files it can scan
- Word
- Excel
- PowerPoint
- PNG · JPEG
- Text
- Markdown
- HTML
Macros, scripts, formulas and links are read as text only — never opened, run or followed.
Evidence and coverage, not a verdict
Every result shows where each finding is — page, slide, cell or line — and exactly which parts of the file were inspected. If something couldn't be checked, PromptShield says so.
A first check, not a guarantee. PromptShield can miss attacks and flag harmless text. “No threats detected” means no known patterns were found in the parts that were inspected — not that a file is safe.
Private by design
- Files are read on your phone and never uploaded. The Android app has no internet permission.
- You pick each file with your phone's file picker; PromptShield can't see anything else.
- No account, no ads, no analytics, no tracking.
- Scan history is off by default.