> For the complete documentation index, see [llms.txt](https://mana.gitbook.io/manadia/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://mana.gitbook.io/manadia/7.-security-and-trust-architecture.md).

# 7. Security and Trust Architecture

Security and trust are fundamental components of the manadia technical architecture. Since the manadia ecosystem encompasses compute resources, AI models, data, Agent execution, and on-chain interactions, the system must establish security boundaries across multiple layers.

### 7.1 Model Security

Model security includes model weight protection, parameter protection, version management, and access control. For prediction models and third-party Agents, core strategies and parameters should not be arbitrarily copied during execution. TEE, permission management, model hashing, and version attestation collectively enhance model security.

### 7.2 Data Security

Data security includes data encryption, data source signatures, access control, timestamp recording, and data usage governance. Proprietary datasets, paid data, and sensitive factors should never be exposed in plaintext, while maintaining records of which models have accessed specific datasets.

### 7.3 Agent Execution Security

Since AI Agents possess tool invocation capabilities, strict permission controls are required. Different Agents should have different tool permissions, invocation quotas, and execution boundaries. Tools involving wallets, trading, asset management, and on-chain operations require more stringent authorization and confirmation mechanisms.

### 7.4 Audit and Replay

The auditing system should support replaying critical execution processes. Replay does not necessarily require revealing proprietary model logic, but it should be able to verify the consistency among the model version, input data digest, strategy configuration, execution timestamp, and output results.

### 7.5 Trust Boundaries

The Trustworthy AI Prediction Model must clearly define its technical boundaries. AI prediction does not imply deterministic outcomes. TEE does not guarantee investment returns. Historical performance does not guarantee future results. Explicitly defining these boundaries within the white paper strengthens technical credibility rather than diminishing the project's value.