PROMPT INJECTION CHECK

Methodology & testing

Prompt Injection Check is a browser-based preflight scanner for text and supported documents. Scanned content is analyzed locally in your browser using deterministic, rule-based checks. No AI model is used to classify scanned content, and scanning does not require document upload.

What the scanner does

The scanner looks for hidden-content patterns, instruction-like signals, obfuscation and document structures that may deserve review before content is given to an AI system.

Findings and coverage

A finding means the scanner observed a signal worth reviewing. It does not prove that a document is malicious. Coverage shows what the scanner was able to inspect and helps distinguish "nothing found" from "not fully inspected".

Testing and regression

Detector changes are checked against maintained regression sets before release. Testing tracks true positives, false negatives, false positives and true negatives, while the detector engine version is kept separate from the website version.

Versioning and reproducibility

Site updates and detector updates are versioned separately. Release packages are hashed with SHA-256, and previous validated releases are retained for rollback.

Privacy

Scanned content is processed in your browser. Prompt Injection Check does not require an account or document upload for scanning.

Limitations

Prompt injection detection is heuristic by nature. Obfuscated or previously unseen techniques may be missed, and benign content may occasionally be flagged. Results support review; they are not a security guarantee.

Project transparency

Prompt Injection Check is an independent project maintained by AdrianRepair. Public documentation explains the product, methodology, limitations and release history so changes can be reviewed over time.

Release metadata

Site version: v90.5
Detector engine: v70-2026-09-29