Security & Governance Architecture
This section describes how the protocol secures user funds, manages restaked assets, governs AI decision-making, and ensures safe protocol upgrades through transparent, on-chain mechanisms.
The architecture is designed specifically for a Restaking-Powered AI Infrastructure on Solana, where economic security, AI-assisted risk management, and governance separation are core principles.
Design Principles
The protocol security model is built on five core principles:
Restaking-backed economic security
AI-assisted risk detection, not fund control
Strict separation of roles and authorities
Multisig-first asset custody
Governance-guarded upgrades with fail-safes
Smart Contract Security Architecture
Modular Protocol Design
The protocol is composed of isolated smart contract modules:
Restaking Vault Module Handles user deposits, withdrawals, and issuance of restaked positions.
Delegation & Allocation Module Allocates restaked assets across validators, services, or AI-secured tasks.
AI Risk Engine Interface Consumes off-chain AI risk scores and feeds them into on-chain risk parameters.
Slashing & Risk Control Module Applies exposure limits, cooldowns, and automated de-delegation.
Governance & Authority Module Controls upgrades, parameters, and emergency actions.
Each module is permission-scoped to minimize blast radius in case of failure.
Core Security Invariants
The protocol enforces the following invariants at all times:
Users cannot withdraw more assets than their restaked balance.
Delegation cannot be changed without valid on-chain authorization.
AI systems cannot directly move or control user funds.
All protocol upgrades require multisig approval.
Treasury assets are isolated from protocol execution logic.
Squads Multisig Treasury & Asset Custody
Squads as the Asset Custody Layer
All protocol-owned assets are secured using Squads Multisig, the standard multisig infrastructure on Solana.
Squads is used for:
Protocol treasury custody
Restaking reward management
Insurance & risk reserve funds
Operational and ecosystem allocations
No single-signer wallet is used to custody protocol funds.
Multisig Configuration
Purpose
Multisig Setup
Treasury & Asset Custody
Squads Multisig 3/5
Protocol Upgrade Authority
Squads Multisig 3/5
AI Parameter Updates
Squads Multisig 2/3
Emergency Guardian
Limited authority (pause-only)
Key Separation & Operational Security
To reduce operational risk:
Treasury keys are separated from upgrade keys.
All multisig signers use hardware wallets.
No multisig signer has unilateral execution power.
All executed transactions are publicly verifiable on-chain via Squads.
Restaking Risk & Slashing Protection Layer
Slashing Risk Model
Restaked assets may be exposed to slashing due to:
Validator misbehavior (double signing, prolonged downtime)
Service-level failures
Risk threshold violations detected by AI monitoring
Protocol-defined economic penalties
AI-Assisted Risk Mitigation
The AI system acts as a risk intelligence layer, not a custodian:
Continuously monitors validator performance and on-chain signals
Generates risk scores for validators and services
Triggers automated safeguards when thresholds are exceeded
Safeguards include:
Exposure caps per validator or service
Automatic rebalancing or de-delegation
Cooldown periods before stake reallocation
Early warning alerts for governance intervention
AI Does Not Control Funds
The AI Risk Engine:
Cannot sign transactions
Cannot withdraw, transfer, or reallocate funds directly
Cannot bypass governance or multisig controls
All fund movements are executed exclusively by smart contracts governed by multisig-approved parameters.
AI Governance & Model Security
AI Model Update Governance
AI model updates follow a controlled process:
Model training and validation off-chain
Model version published to IPFS / Arweave
Model hash registered on-chain
Squads multisig approval required
Activation after a predefined delay window
This ensures transparency, auditability, and protection against malicious updates.
Anti-Manipulation & Data Integrity
To prevent AI manipulation:
Risk signals are sourced from multiple independent data feeds
On-chain behavior is cross-validated
Anomaly detection flags outliers
Risk score outputs are bounded within protocol-defined limits
All AI decisions affecting protocol parameters are publicly observable.
Governance & Protocol Upgrades
Governance Scope
Governance controls:
Protocol upgrades
Risk parameters and thresholds
Treasury allocation
AI integration rules
Emergency response actions
Upgrade Safety Mechanisms
All upgrades require:
Multisig proposal and approval
Public notice and review period
Optional timelock (24–72 hours)
Post-upgrade monitoring
Upgrades cannot:
Mint new tokens outside tokenomics
Bypass slashing rules
Access treasury funds without multisig approval
Emergency Controls
In case of critical risk:
Emergency guardian can pause protocol operations
Pausing does not allow fund withdrawal
Resumption requires multisig confirmation
All emergency actions are publicly logged
Audit & Verification Strategy
The protocol follows a multi-layer security review process:
Independent smart contract audits (OtterSec, Halborn, Zellic, Certik )
Continuous testing and fuzzing
AI risk simulation and adversarial testing
Ongoing monitoring post-deployment
Last updated