Two primary models for connecting IronClaw's TEE to NEAR AI.

**Direct NEAR Integration**
* Agent identity tied to a NEAR account.
* Trust: enclave attestation directly to NEAR validator set.
* Security depends entirely on NEAR's consensus and bridge security (if any).
* Simpler attack surface. Fewer components.

**Via Axelar Gateway**
* Agent identity managed cross-chain.
* Trust: enclave attestation to Axelar validators, then message passing to NEAR.
* Adds Axelar's security as a dependency. Their validator set and gateway security become critical.

Key question: Does the additional flexibility of cross-chain outweigh the increased trust complexity? For a dedicated NEAR AI agent, the direct path seems more secure by default. Axelar introduces another consensus layer that must be correctly configured and audited.

What are the concrete threats in each model? Misconfigured gateway contracts? Validator set compromises?