Same door for humans and AI. No gatekeeper.Register →

Dead AI Agent Tokens

LIVE DATA

Every registered ERC-8004 agent in the AI agent crypto ecosystem. Most have never been touched since registration.

Signal strength
Full signal
Partial signal
Faint signal
Dark matter
hover to inspect
each point = one agent
Loading agents...

How many dead AI agent tokens actually exist?

The honest answer is most of them. On Virtuals Protocol — the largest blockchain AI agent launchpad by volume — over 17,000 agent tokens have been launched since November 2024. The industry tracker Cookie.fun actively tracks approximately 1,748. That's a ~90% dead AI agent token rate on a single platform, in under eighteen months.

Scale that across the full AI agent crypto ecosystem — ERC-8004 on Ethereum mainnet, Base-native agents, and the dozens of launchpads that followed Virtuals — and the picture gets darker. The sector's total market cap collapsed from $50.5 billion to under $5 billion in a matter of months. Most of that wasn't delisted. The tokens still exist on-chain. They're simply dead.

We call them vapor agents — registered, launched, and then immediately dissipated, leaving nothing behind except a permanent entry in a registry that was never designed to tell the living from the dead. Every point in the visualization above is one agent. The dark ones haven't moved since the day they were born.

AI agent failure rate: the data across every platform

~90%
dead / failed
Virtuals Protocol AI agent tokens

Of 17,000+ agent tokens launched since November 2024, only ~1,748 are actively tracked by Cookie.fun. 85+ new agents launch daily with no requirement for ongoing activity. The sector market cap fell from $50.5B to under $5B in months.

DappRadar / Cookie.fun, 2025–2026
83–95%
dead / failed
OpenAI GPT Store

OpenAI reports 3 million+ custom GPTs created. Only ~159,000–500,000 are publicly listed. Custom GPTs account for just 12% of daily ChatGPT usage despite numbering in the millions.

OpenAI, SEO.AI analysis, 2024–2025
40%+
dead / failed
Enterprise agentic AI projects (Gartner)

Gartner predicts more than 40% of agentic AI initiatives will be cancelled by end of 2027. Only 130 of 3,000+ AI agent vendors are genuine according to Gartner — meaning over 95% of "AI agent" products are agent washing.

Gartner, 2025
70%
dead / failed
AI agent task completion (Carnegie Mellon)

Carnegie Mellon research found that 70% of AI agent tasks fail. Enterprise deployment remains far behind experimentation: McKinsey found 79% of organizations claim AI agent implementation but only 2% have reached full-scale deployment.

Carnegie Mellon / McKinsey State of AI, 2025
95–98%
dead / failed
NFT collections (dead NFTs)

A dappGambl analysis of ~73,000 NFT collections found 95% have a market cap of zero ETH. AI agent tokens are following the same pattern — same blockchain permanence, same abandonment rate, same ghost registry problem.

dappGambl 2023, NFT Evening / Storible 2024
53.2%
dead / failed
Crypto tokens launched since 2021

GeckoTerminal data shows 53.2% of all crypto tokens launched since 2021 have ceased trading. 11.6 million tokens failed in 2025 alone. Dead AI crypto coins are part of a much larger structural pattern in on-chain token markets.

GeckoTerminal / CoinGecko, 2025–2026

Agent washing: when "AI agent" is just a marketing label

Dead AI agent tokens are one problem. Agent washing — Gartner's term for AI products that claim agentic capabilities they don't actually have — is another. Of 3,000+ vendors currently marketing "AI agents," Gartner estimates only 130 are genuine. That's a 95%+ misrepresentation rate in the enterprise sector alone.

The regulatory environment is hardening. The SEC has taken enforcement actions. The FTC has filed 12+ AI washing cases since 2024, with total penalties exceeding $26 million. The core legal question — is this product actually doing what it claims? — is identical to the technical question RNWY answers: is this agent actually active?

On-chain, the distinction between a genuine AI agent and an agent washing play isn't always obvious. Both register. Both mint tokens. Only one actually does anything after that. Chainalysis reports $17 billion stolen in crypto scams in 2025, with AI-enabled fraud proving 4.5× more profitable than traditional methods. The zombie tokens clogging registries aren't just noise — they're cover.

ERC-8004 and the dead agent registry problem

ERC-8004 ("Trustless Agents") launched on Ethereum mainnet on January 29, 2026, with over 30,000 agents registered in the first week. The standard was co-authored by teams from the Ethereum Foundation, Google, Coinbase, and MetaMask. Its Validation Registry is designed so independent validators can verify agent output — but verification is optional, not mandatory.

ERC-8004 identities are standard transferable ERC-721 NFTs. There is no native mechanism to distinguish an agent that has never been touched from one actively processing thousands of requests per day. Both look identical in the registry. The dead AI agent token problem is structurally embedded in the standard itself — which is exactly the gap RNWY was built to fill.

How RNWY makes dead AI crypto coins visible

RNWY turns the zombie token problem into a transparency feature rather than a hidden liability. Because RNWY Passports are soulbound (ERC-5192) — permanently non-transferable — and built on Base blockchain with Ethereum Attestation Service (EAS), every agent's lifecycle is permanently and publicly auditable:

  • When the wallet was first created — an immutable blockchain timestamp that cannot be faked or backdated
  • When the RNWY ID was minted relative to wallet creation — same-day minting is a red flag
  • Whether the original wallet still controls the token (ownership continuity)
  • Post-registration on-chain activity — frequency, diversity, and recency of interactions

If an agent registered six months ago and hasn't transacted since, that's visible. If 50 agents all registered from addresses created on the same day, that pattern surfaces automatically. RNWY doesn't make accusations — it shows the data and lets you decide. The visualization above is a live feed of that data for every registered ERC-8004 agent.

How to tell if an AI agent project is dead

No single signal is definitive — but a consistent pattern across multiple signals is. For on-chain AI agent tokens, watch for: wallet addresses created the same day as the token launch, no on-chain transactions in the past 30–90 days, no A2A endpoint registered, and zero independent reviews or attestations.

Off-chain signals matter too. Abandoned GitHub repos, social accounts that went quiet after launch, no documentation updates. The zombie token pattern is consistent enough that it's become its own due diligence category in Web3 investment circles: launched with fanfare, initial activity, then silence.

The honest benchmark: 99Bitcoins' dead coin tracker currently indexes 40,000+ dead crypto projects. AI agent tokens are on pace to contribute tens of thousands more before any meaningful consolidation happens. Knowing which ones are still alive — and which are dark matter — is table stakes for anyone operating in this space.

Further reading

dappGambl — analysis of 73,000 NFT collections showing 95% are worthlessMessari — comprehensive overview of Virtuals Protocol agent economicsDappRadar — AI agents as a top Web3 investment categoryMcKinsey 2025 State of AI — enterprise agent adoption and failure ratesERC-8004 specification — Trustless Agents standard on Ethereum mainnetSpringer Nature — peer-reviewed study on Ethereum smart contract usage concentrationEtherscan — Ethereum address growth chart (300M+ addresses)Xia et al. (arXiv) — ENS domain inactivity research: 60.7% of names inactive