How it works

Local prompt risk detection before data is sent to AI tools

Trust-Prompt adds a local precheck layer between you and AI chat tools. Before a prompt is sent, it is analyzed in your browser — deterministic, offline, and enforced before transmission.

StoreBasic: ChatGPT-only scope Pro: multi-site scope + upload policy + audit + settings lock
Install StoreBasic (Chrome Web Store) Pricing Testing & Quality Privacy Policy
Simulation · Extension panel (toggle StoreBasic ↔ Pro)
StoreBasic
Trust-Prompt
ChatGPT-only · Local checks
Scope: ChatGPT
Demo

Status

Checking…
This is a website simulation. StoreBasic enforces checks locally on ChatGPT.

Status

Panel controls are a website demo. They do not change your real extension settings.

Controls

Protection
Toggle on/off (simulation only). Real extension enforcement is not affected.
Upload policy
Scope controls
Pro controls are disabled in StoreBasic mode (simulation). Toggle to Pro to preview Pro-only modules.
Try it: toggle Pro, press Run demo check, then tap WARN or BLOCK. Scroll inside the panel to see everything (Audit + Upload sections are further down on this page).

StoreBasic is ChatGPT-only. Pro adds scope controls, upload policy, settings lock, and a local audit view when licensed.

Step 1 – Intercept before sending

Trust-Prompt reliably holds the send action until the decision is known (pre-send enforcement).

StoreBasic intercepts prompt submissions on supported sites (ChatGPT). Trust-Prompt prevents sending until the check finishes — so the decision happens before transmission.

StoreBasic applies here Pro also applies
Simulation · Precheck states (StoreBasic + Pro)
Checking…
Example prompt
“Can you help me summarize this contract clause?”

Warning — review before sending

This message contains medium-risk indicators. Proceed only via explicit confirmation.

Proceed anyway Edit message

Simulation for explanation. Real enforcement applies in StoreBasic and Pro.

Trust-Prompt warning overlay requiring explicit user confirmation before sending

Real UI example: WARN requires explicit user confirmation. (StoreBasic + Pro)

Step 2 – Local rule-based analysis

Signals are detected locally (offline) and mapped to WARN/BLOCK decisions with a deterministic ruleset.

In StoreBasic, prompts are evaluated locally with a versioned ruleset: no servers, no cloud processing, no AI calls. Pro uses the same deterministic approach and adds Pro-only modules around it.

StoreBasic: deterministic local rules Pro: deterministic rules + Pro modules
  • Financial data (e.g., valid IBANs verified via MOD97) — BLOCK
  • Payment card combinations (card + CVV / card + expiry) — BLOCK
  • Secrets and tokens (API keys, JWT-like tokens, private keys) — BLOCK
  • PII indicators (email, phone number, address-like patterns) — WARN
  • Upload event detection (attachment added) — no file inspection (StoreBasic + Pro)

Step 3 – Clear decision (WARN / BLOCK)

Two enforced outcomes: WARN (confirm to proceed) and BLOCK (send prevented).

Trust-Prompt surfaces a decision before sending: BLOCK prevents transmission; WARN requires explicit consent to proceed.

StoreBasic: enforced WARN/BLOCK Pro: enforced WARN/BLOCK
Trust-Prompt warning overlay allowing review and explicit proceed

WARN example (StoreBasic + Pro).

Trust-Prompt block overlay preventing prompt submission when high-risk content is detected

BLOCK example (StoreBasic + Pro).

Audit (Pro)

Pro adds a local audit view for governance signals (counts + rule IDs) without storing prompt content.

Pro only: A local audit view can help governance by showing usage signals (e.g., checks, warnings, blocks, and user decisions) without storing prompt content.

Pro-only feature: local audit (no prompt content)
Audit
Local audit trail (no prompt content)
Local

Checks

93
Total evaluations

Warnings

18
User review needed

Blocks

36
Send prevented

Proceed / Stop / Redact

0 / 13 / 5
User decisions

Awareness score

0
Higher = safer behavior
Events
72
Showing WARN/BLOCK/PROCEED/STOP/REDACT (ALLOW hidden)
WARN BLOCK PROCEED STOP REDACT

Audit panel is a simulation for the website. Audit is Pro only.

File uploads

StoreBasic never reads files. It only detects an attachment event and warns. Pro can enforce upload policies.

StoreBasic: no file inspection, no OCR, no document analysis. Trust-Prompt only detects the attachment event and warns users to avoid uploading sensitive documents to AI tools.

Pro: adds upload policy controls (allow / warn / block) for stricter governance.

StoreBasic: upload event warning only Pro: upload policy controls
Trust-Prompt upload warning shown when an attachment is added (no file inspection)

Upload event warning (StoreBasic + Pro). File content is not read.

Pro controls (interactive preview)

Click the items below to preview what Pro modules do (website demo).

Pro only: modules help protect settings and enforce policies in stricter environments. Click below to preview what each Pro control does.

Pro-only: settings lock + upload policy + scope controls + local audit
Simulation · Pro modules
Licensed Pro
Settings Lock
Protect Pro settings from tampering (Pro only)
Pro
Upload Policy
Allow / Warn / Block upload events (Pro only)
Pro
Scope Controls (Multi-Site)
ChatGPT only / All supported / Custom (Pro only)
Pro
View Pricing Testing & Quality Contact / Support

Privacy by design (GDPR / EU AI Act alignment)

Designed to minimize data exposure: local checks, no telemetry, and no prompt storage by default.

  • No telemetry
  • No logging of prompt content
  • No external requests for prompt checking (StoreBasic)
  • No user profiling

Trust-Prompt follows privacy-by-design and data-minimization principles aligned with GDPR/DSGVO expectations. It is a technical safeguard and does not replace legal advice or internal compliance processes.

Official references: GDPR (Regulation (EU) 2016/679) — EUR-Lex · EU AI Act (Regulation (EU) 2024/1689) — EUR-Lex · European Commission — Regulatory framework for AI