4.1 The Functional Prototype: SecretML v4
The current reference implementation, SecretML v4, serves as a stable, cross-platform proof of concept (optimized for Windows 10+). It features a dedicated graphical user interface (GUI) designed to facilitate secure key generation without compromising Zero-Knowledge integrity.
Core Operational Principles:
- Zero-Persistence State: The application operates in a strictly volatile environment. No sensitive data is ever cached to local databases or external servers.
- Memory Isolation: All critical contexts — questions, answers, and secret payloads — exist only within RAM for the duration of the active session.
- Active Sanitization: Upon completion or manual reset, the
reset_user_data() routine purges all cryptographic material from memory.
4.2 The Key Derivation Pipeline
SecretML v4 utilizes a multi-stage pipeline to forge a 256-bit Master Key. The system supports two primary derivation modes to adapt to different threat levels:
4.2.1 Linear Entropy Fusion (Standard)
The baseline protocol optimized for speed. It layers cognitive responses with a physical file hash into a single, memory-hard derivation cycle.
Kcombined = H(Ans0 + Saltfile) + ∑i=1n H(Ansi)
Kfinal = H(Kcombined)
Where H represents the SHA-256 hashing function.
4.2.2 Phantom-Step Cascade (High-Assurance)
An advanced engine based on Sequential Dependency. In this mode, the system realizes a step-wise decryption cascade: data for Step N does not exist in a readable state until Step N-1 is successfully unlocked.
k0 = Hash(SecretFile)
ki = Argon2id( Answeri, Salt = ki-1 )
Kfinal = Argon2id( kn, Salt = Internal_Pepper )
Each step i acts as the encryption key for the next prompt's metadata.
Technical Advantages:
- Parallelization Resistance: Since step i requires the output of i-1, attackers cannot distribute the workload across GPU cores.
- Chain Integrity: Any single error in the sequence renders the final payload permanently inaccessible, providing a robust defense against forensic analysis.
4.3 AI-Enhanced Security: "Cognitive Assistance"
SecretMemoryLocker integrates Large Language Models (LLMs) to overcome the "Creative Block" of security questions.
- AI-Curated Database: v4 includes 1,000+ pre-generated questions categorized by difficulty and theme.
- Dynamic Question Forge (Upcoming): A local LLM assistant to help users craft highly personal, high-entropy questions that are easy to remember but impossible for others to guess.
4.4 Cryptographic Primitives & Stack
We rely exclusively on industry-standard, audited libraries for the reference implementation:
| Component |
Primitive / Library |
Purpose |
| KDF |
Argon2id |
Memory-hard deterministic key generation. |
| Hashing |
SHA-256 |
Integrity verification and entropy stacking. |
| Encryption |
ChaCha20-Poly1305 |
Authenticated encryption (AEAD) for payloads. |
| Interface |
Tkinter / ttk |
Lightweight, native GUI for Windows environment. |
4.5 Development Roadmap: Hardening & Expansion
SecretML v4 is a concept validation tool. The roadmap for production-grade security includes:
Software & Audit:
- Memory Hardening: Porting core logic to Rust to eliminate Python's memory management vulnerabilities.
- MirageLoop: Automating the generation of "decoy" sequences to misdirect automated tools.
Entropy Anchor Evolution:
- Biometric Integration: Replacing file salts with non-reproducible 256-bit vectors from FaceID, TouchID, or Iris scans.
- Hardware Tokens: Native support for YubiKey (U2F) and e-ID signatures as primary physical anchors.
Security Stance:
The architecture ensures that raw biometric data or hardware secrets are never stored. Only their one-way cryptographic representations enter the hashing pipeline.