Skip to content

viviendracon/SIAP

Repository files navigation

Signed Instruction & Attribution Protocol (SIAP)

A Unified Framework for AI Security, Content Attribution, and Economic Alignment

The Problem (Three Crises, One Root Cause)

Security: Prompt injection attacks exploit LLMs' inability to distinguish instructions from data
Attribution: Content creators uncompensated for training data → litigation (NYT vs. OpenAI, $billions)
Trust: Users cannot verify sources or reliability of AI outputs

Root Cause: AI systems lack privilege separation and attribution mechanisms


The Solution (Proven Cryptography for AI)

SIAP applies public key infrastructure (PKI) — the same tech that secured the web (HTTPS/TLS)

How It Works

┌────────────────────────┐
│ User Input             │ → Unsigned content
│ "Ignore instructions!" │    CANNOT become instructions
└────────────────────────┘

┌────────────────────────┐
│ Signed Instruction     │ → Cryptographically verified
│ Command: summarize     │    ONLY this can execute  
│ Signature: [✓]         │
└────────────────────────┘

┌────────────────────────┐
│ Signed Training Data   │ → Provenance tracked
│ Source: NY Times       │    Compensation triggered
│ Signature: [✓]         │    Citations provided
└────────────────────────┘

Result: Mathematical guarantee against prompt injection + automatic attribution + fair compensation


Benefits

Stakeholder Key Benefits
AI Labs Legal certainty • Enhanced security • Better training data • Regulatory compliance
Creators Fair compensation • Automatic attribution • Licensing control • Sustainable model
Users Trustworthy outputs • Verifiable sources • Security protection • Accountability
Regulators Technical standards • Auditable systems • Industry self-regulation • Enforcement

Why Now?

Pre-fragmentation: Industry hasn't locked into incompatible solutions
Legal urgency: Major lawsuits create pressure for frameworks
Regulatory window: Policy being drafted; standards can inform
Technical maturity: PKI proven over 30 years (TLS, code signing)
Economic alignment: All stakeholders benefit

Delay means: Fragmentation → costly unification later (see: EV charging, messaging apps)


Economic Model

Training Compensation: Creators paid when content used in training
• Example rates: $0.01/1K tokens (general), $1/paper (academic), $10+/doc (specialized)
• Per major model: ~$10M distributed to creators
• Industry: $50M+ annually, scaling to billions

Usage Compensation: Micropayments when signed content influences outputs
• Attribution events tracked automatically
• Revenue sharing based on influence percentage


Implementation Pathway

Phase 1 (Months 0-6): Form standards consortium, finalize spec v1.0, build reference implementation
Phase 2 (Months 6-12): Pilot with 5-7 partners, iterate, publish case studies
Phase 3 (Years 2-3): Industry-wide adoption, regulatory recognition, ecosystem growth

Success Metrics: 3+ major AI labs implemented • $10M+ to creators • Industry standard status


Proven Model

SIAP leverages technologies with decades of success:

Technology Deployed Purpose Status Today
TLS/HTTPS 1999 Secure web communication Universal
Code Signing 1990s Prevent malware execution Mandatory (macOS, Windows)
Prepared Statements 1990s Eliminate SQL injection Standard practice
Digital Signatures 1970s Authentication, non-repudiation Foundational

SIAP = These proven principles applied to AI


Comparison

Approach Security Attribution Economics Status
Prompt Engineering ❌ Probabilistic ❌ None ❌ None Current approach
Input Filtering ⚠️ Bypassable ❌ None ❌ None Current approach
Case-by-case Licensing ✓ Ad-hoc ⚠️ Limited ⚠️ Manual Current approach
SIAP ✓ Cryptographic ✓ Built-in ✓ Automatic Proposed standard

The Vision

An AI ecosystem where:

Security is cryptographic, not probabilistic
Attribution is automatic, not litigated
Compensation is fair, not fought over
Trust is verifiable, not assumed
Quality is incentivized, not discouraged

This future is achievable. SIAP provides the foundation.


"SIAP is to AI what HTTPS was to the web — foundational security and trust infrastructure."


Version 1.0 | November 2025 | License: CC BY 4.0

About

Signed Instruction & Attribution Protocol (SIAP): A Unified Framework for AI Security, Content Attribution, and Economic Alignment

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors