How Does Giza Work?
Giza is an autonomous AI agent protocol that lets users deploy non-custodial AI agents to optimize DeFi yield strategies across multiple chains. Its flagship product ARMA automatically reallocates stablecoin deposits across lending protocols like Aave, Compound, and Morpho. With $21M in TVL and $5.7M in funding, Giza receives a C grade reflecting the novelty risks of AI-driven autonomous execution combined with the absence of formal security audits.
TVL
$6M
Sector
DeFi
Risk Grade
C+
Value Grade
C-
Core Mechanisms
Governance > Agent-Based Execution
NovelAutonomous AI agents (ARMA) execute DeFi strategies without human intervention using verifiable ML models
Novel approach to on-chain ML verification for autonomous trading
Custody > Session-Key Authorization
NovelSmart Authorization Layer uses session keys to allow agents to act on behalf of users without full custody transfer
Self-custodial agent execution via scoped session keys
Yield > Strategy Optimization
ARMA agent continuously monitors and reallocates stablecoin deposits across Aave, Compound, Morpho, Moonwell for optimal yield
Yield optimization strategy across multiple lending protocols
Data > Semantic Layer
NovelTranslates DeFi protocol operations into agent-readable structured data for autonomous decision-making
Novel semantic abstraction layer for AI-DeFi interface
Token > Governance & Utility
GIZA token used for governance voting and 20% revenue buyback mechanism
Standard governance token with buyback utility
Execution > Decentralized Agents
Decentralized Execution Layer carries out agent instructions across multiple chains (Base, Mode, Arbitrum)
Multi-chain execution infrastructure for agent operations
How the Pieces Interact
Agent may execute strategies that interact with vulnerable smart contracts or exploit unintended protocol behaviors
Stale or manipulated price feeds could cause agents to make suboptimal allocation decisions
Compromised session keys could allow unauthorized agent actions on user funds
Exploit in any integrated protocol could cascade through agent-managed positions
Insufficient protocol revenue could undermine buyback sustainability and token value
What Could Go Wrong
- AI agent execution introduces novel autonomous decision-making risks where agents may execute suboptimal or harmful strategies without human oversight
- No formal security audits reported despite managing user funds through session-key authorization and smart contract interactions
- Multi-protocol exposure through yield optimization strategies creates cascading risk if any integrated protocol (Aave, Compound, Morpho, Moonwell) experiences an exploit
AI Agent Strategy Failure Cascade
ModerateTrigger: ARMA agent executes a strategy that interacts with a compromised or exploited integrated protocol, causing significant losses across managed positions
- 1.Integrated protocol (e.g., Morpho or Moonwell) experiences a smart contract exploit — Agent-managed funds allocated to that protocol are at risk of loss
- 2.ARMA agent detects anomaly but reallocation logic moves funds to another stressed protocol — Losses compound as agent follows programmed strategy without human judgment
- 3.Multiple users' agents simultaneously attempt to exit positions — Liquidity crunch in integrated protocols as automated withdrawals spike
- 4.GIZA token price drops as confidence in agent reliability falls — Buyback mechanism becomes less effective, further reducing token demand
- 5.Institutional partners (Re7 Capital) withdraw managed capital — TVL decline accelerates, reducing protocol revenue and sustainability
Risk Profile at a Glance
Overall: C+ (41/100)
Lower score = safer