Overview

SecureGPT is a privacy-first AI application designed to ensure your data stays protected during AI interactions. It utilizes advanced preprocessing techniques, such as Named Entity Recognition (NER) models, to automatically filter sensitive information before it reaches the cloud. With SecureGPT, you can interact with powerful AI models like ChatGPT and Gemini securely, using customizable filters and optional anonymous routing (TOR) for ultimate privacy.

Modes of Operation

The SecureGPT dApp operates in the following modes:

1. Secure Mode
   • All queries are preprocessed through our NER models to filter sensitive information.
2. Normal Mode
   • Preprocessing is skipped, and you can access online models like OpenAI and Gemini as usual.
3. Anon Mode
   • TOR mode is activated. All queries route through TOR-enabled sockets, providing ultimate privacy for your interactions.

Filter Types

1. Default
   • All critical, identified information will be filtered automatically. Users
     cannot choose what to expose or filter.
2.  Custom
   • (Selective Data Guard) Users can select specific information to be filtered. For example, in a fitness AI agent, you might choose to retain personal information like age, sex, height, weight, and BMI, but filter out non-essential information.

Phase 1: PII and PCI Models

• Secure Mode Only. Default Filter.
  What is PII and PCI?
• PII (Personally Identifiable Information) includes your name, phone number, address, DoB, Social Security Number, etc.
• PCI (Payment Card Industry) includes sensitive payment data like credit/debit card details.

Most common leaks in AI interactions are personal or payment-related information, often exposed unintentionally. Consider the following examples:

• Example 1:
  Your boss asks for help drafting an email. You paste the email into an AI interface. Without realizing it, you’ve shared personal or sensitive work related details.
• Example 2:
  While preparing your resume using an AI assistant, you inadvertently expose sensitive information about your personal history.

These leaks can happen without users realizing it, which is why Phase 1 focuses on safeguarding personal and payment-related data.

Phase 2: OpenAI, Gemini, PII, and VoiceGuard

Secure Mode and Normal Mode. Default Filter.
In this phase, we will introduce VoiceGuard, which allows voice commands
and transcriptions to be processed securely. Users can toggle between
OpenAI and Gemini models, both with privacy filters in place.

Phase 3: PHI Models and Extended Modes

• Secure Mode, Normal Mode, and Anon Mode. Default Filter.
  PHI (Protected Health Information) includes data like health conditions, drug usage, injuries, and blood types. As part of HIPAA compliance, organizations must ensure that sensitive health information is redacted before transmission. This phase introduces PHI models, enhancing privacy and compliance for sensitive health-related information.

Phase 4: Custom Filters + All Models + All Modes

In Phase 4, Custom Filters will be fully operational, allowing users to specify exactly what information to redact or retain. All models and modes will be available, allowing for complete customization of the filtering process based on your specific needs

Using SecureGPT:

1. On the home screen, connect any fresh wallet. For this testing phase, there are no specific requirements. Just connect a fresh wallet.
2. On the landing page, click SecureGPT.
3. For Phase 1, only OpenAI model is available. Click OpenAI.
4. You will land on the chat screen. Chat as you would with any other AI app.s
5. Click the bar on the top left side of the screen to see your chats.
   
6. The chats are available on the left side of the screen.
7. To see how your query is preprocessed, click the button as shown below.
   
8. After clicking the button, the following window will appear, showing the preprocessed information.