Eliza’s Token Is Dead. Can the Framework Survive Without It?

On August 4, the developer who built one of crypto’s most widely used artificial-intelligence projects wrote its obituary himself. In a long message on X, Shaw Walters, the founder of Eliza Labs, told the market what many holders had already priced in. “The token is dead. Completely. The foundation is winding down,” he wrote, a statement picked up within hours by CoinDesk, The Block and Decrypt.

That one line closed a story that began as a joke and ended in federal court. It did not close the project. On the same day the token was pronounced dead, the code kept moving: the elizaOS GitHub repository logged fresh commits, its star count sat unmoved near 19,000, and a publicly traded company was still on schedule to ship a commercial product built on the framework within days. For a sector that has spent two years insisting a token is a project and a project is a token, Eliza’s split ending is the cleanest test yet of whether an open-source framework can outlive the speculative asset that funded it.

This is an account of that test. What Walters actually announced, how a $2.39 billion market cap became a rounding error, why a private lawsuit rather than the SEC delivered the final blow, and the question crypto rarely has to answer: who keeps maintaining a widely adopted open-source framework once its foundation dissolves and its token goes to zero?

What Shaw Walters actually announced

Walters did not soften it. Alongside the death notice he explained the mechanics. Eliza Labs and its foundation had settled the class action filed against them, and they had done so by handing over essentially everything they had left. “Their claim was ridiculous, but we didn’t have the capital to legally fight it so we settled on giving them the rest of what we had,” he wrote, in comments reported by The Block. The settlement consumed the foundation’s remaining treasury and available cash, according to Decrypt.

Then came the part that matters to everyone still building on Eliza. Walters said he would keep developing the software, but never again attach a token to it. “I am starting over, since I own the IP, and I am never letting a token come close to Eliza again,” he wrote. He was blunt about his own position: “I don’t own any tokens. I don’t support any of it.” According to CoinDesk, Walters had once held roughly $25 million worth of the token; that stake has fallen toward nothing along with everyone else’s.

He framed the reset as continuity rather than surrender. “Gonna keep building no matter what, every single day,” he wrote, and, in the line that spread fastest, “Eliza is dead. Long live Eliza.” The distinction he was drawing is not rhetorical. It is legal and structural. The token, the foundation and the treasury are being wound down. The framework, an MIT-licensed body of TypeScript that anyone can copy, run and fork, is not his to kill even if he wanted to.

From a $2.4 billion peak to a rounding error

The arc is short and steep. Eliza launched in October 2024 as ai16z, a play on the venture firm Andreessen Horowitz, spun up on Solana through the daos.fun launchpad with a modest five-figure funding target. It caught fire when Marc Andreessen himself posted a public shoutout, a moment that pushed the market cap into the tens of millions of dollars within hours and briefly knocked the launchpad offline. By January 2, 2025, the token peaked at around a $2.39 billion market cap, according to CoinDesk.

Everything after that was decline. Andreessen Horowitz asked the project to stop leaning on its name, and in late January 2025 ai16z rebranded to ElizaOS. A token migration in November 2025 swapped the old AI16Z supply for a new ELIZAOS one. Neither move arrested the slide. As of this writing the token trades near $0.00023 for a market cap of roughly $1.75 million, having printed a fresh all-time low on August 8, according to CoinGecko. That is a fall of more than 99% from the peak.

The table makes the shape obvious. Nearly all of the value was created and destroyed in the first quarter of the project’s life, long before the lawsuit, and the framework’s technical progress ran on a completely separate clock.

The autonomy gap, the lawsuit, and a settlement of last resort

The claim at the center of the whole affair was always about autonomy. ai16z was marketed as an AI-run venture fund whose agents made the calls. Reporting from Protos at the end of October 2024 punctured that framing, describing a setup where humans still approved the trades. Walters himself grew more candid over time; in an earlier interview with Decrypt he conceded, “You probably do not want to give an AI agent a bunch of money and expect it to make you more.” The gap between the pitch and the practice is exactly what a plaintiff’s lawyer needs.

The suit Walters settled is Pikabea v. Walters, filed in the Southern District of New York on April 20, 2026 by Burwick Law on behalf of a Spanish holder, Gorka Pikabea. The federal docket lists Eliza Labs, Walters, ElizaOS director Sebastian Quinn-Watson and the AI16Z DAO among the defendants. The complaint alleged false advertising, deceptive business practices, negligent misrepresentation and unjust enrichment, and it took aim at the token migration itself, alleging that the swap steered roughly 40% of the new supply to insiders rather than to existing holders. The proposed class covered thousands of wallets that took net losses over an eighteen-month window.

It is worth being precise about who ended the token. The blow came from private plaintiffs, not a regulator. The SEC never brought a case here. That is closer to a securities-style misrepresentation fight than to the outright exit scams that drain a treasury overnight, and it is a recurring feature of crypto’s 2026 accountability story: where enforcement is slow or absent, class-action firms move first.

The case also showed why suing a decentralized project is hard. By mid-May 2026 it had stalled on the most basic step, serving the defendants. At a hearing covered by Inner City Press, Judge Jed S. Rakoff was unimpressed that plaintiffs had made a single attempt to serve Walters: “I am not going to move the case forward on this basis: 1 attempt?” Quinn-Watson lives in Australia, so serving him required the Hague Convention process, a matter of months. In the end the parties never litigated to a verdict. Walters settled because, by his own account, fighting cost more than the project had left. The allegations were never tested in court, and the defendants admitted no wrongdoing.

Why the framework is not the token, and never was

The reason Eliza can lose its token and keep breathing comes down to a license file. The framework is released under the permissive MIT license, which means the code is not a company asset that evaporates when the company does. Anyone can download it, run it, modify it and ship products with it, with or without Walters, with or without a foundation, with or without a token.

That is why the GitHub numbers barely flinched during the week the token died. The main elizaOS/eliza repository carries roughly 19,000 stars and 5,600 forks, is written in TypeScript, and now bills itself in its own tagline as an agentic operating system rather than a mere agent library. Hundreds of contributors have touched it, and commits landed the same week the obituary did.

Crypto has repeated the phrase framework quality is not token performance for a year without ever seeing the two fully separate. Eliza is now that separation made total: the token is worth essentially nothing while the code is arguably in its most capable state ever. For builders, that is either reassuring or unsettling, depending on what they assumed the token was paying for. It is also the crux of the wider reckoning the agent sector has been having all year over what these projects actually are once the speculation drains out.

Inside Eliza: the architecture that outlived the money

For readers meeting the project through its funeral, a quick orientation. Eliza is a framework for building autonomous AI agents: software that pairs a large language model, the part that plans and interprets, with memory, a set of tools and, frequently, its own crypto wallet, so it can act rather than just answer. That is a different thing from a scripted trading bot, and it is why security researchers treat agents as their own risk category.

Everything in Eliza plugs into a single runtime through one plugin interface. The main moving parts are Actions (things an agent can do), Providers (context an agent can read), Evaluators (post-processing and memory formation) and Services (long-running connections), all registered the same way. Above that sits a messaging model of Worlds (servers or workspaces), Rooms (channels or direct messages) and Entities (users and agents). The design goal is that a developer can add a capability, a data source or an integration by writing a plugin, without touching the core.

That plugin surface is doing double duty in this story. It is why third parties can carry the project forward on their own, and it is also where the security problems live. Both threads run through the rest of this piece. The important point for now is that the value of the framework never sat in the token; it sits in this architecture, the release history and the community that maintains it.

Who maintains a framework after its foundation dissolves?

Here is the situation crypto rarely has to confront. Open-source projects backed by a company usually have a clear answer to who pays the maintainers: the company does, out of revenue or venture funding. Token-funded projects answered it differently, with the treasury and token emissions covering development. Eliza just lost both. The foundation is dissolving and the token cannot fund anything. So who keeps the lights on?

There is no single answer, and that is the point. Three overlapping models are now in play, each with its own incentives and failure modes.

The first steward is Walters himself. He owns the intellectual property, he says he will keep building every single day, and he has reframed his mission around locally run, privacy-preserving agents that give users control of their own data rather than around a tradable asset. A founder-led project with no token and no foundation can be remarkably durable; plenty of significant open-source tools are effectively one person’s life work. It also concentrates risk in burnout, funding and the classic bus factor.

The second steward is the set of commercial adopters, and this is where Eliza’s story diverges most sharply from a typical token collapse. Serious infrastructure teams already run their own copies. Nethermind, one of the major Ethereum client developers, maintains its own fork of the plugin registry. Automata Network runs a registry fork and pairs the framework with hardware-based attestation. These teams do not need the token or the foundation; they need the code to keep working for their products, and they have the engineering capacity to maintain the parts they depend on. The risk is fragmentation: if enough large users maintain private forks with diverging priorities, Eliza stops being one thing.

The third steward is the volunteer community, the hundreds of contributors whose pull requests kept the repository moving. Community maintenance is real and it is also fragile. Without a foundation to triage issues, fund audits and coordinate releases, the coordination overhead falls on whoever volunteers to carry it. The fast release cadence that defined Eliza in 2026 is exactly the kind of thing that is hard to sustain on goodwill alone. The honest answer, days after the announcement, is that Eliza’s maintenance model has become an experiment being run in public.

The commercial afterlife: Agentic SME and the tokenless model

If there is a template for Eliza’s post-token life, it is being built by a small public company most crypto readers have never heard of. In April 2026, Secure Blockchain Development Corp, listed in Canada, completed an all-stock acquisition of Agentic Solutions, a firm founded by ElizaOS director Sebastian Quinn-Watson. In May the two sides signed a development agreement with the Eliza Foundation to build a white-label product called Agentic SME, aimed at helping small and mid-sized businesses automate routine work such as CRM updates and invoice processing, all on top of the open-source framework. The final version targets a mid-August 2026 launch, according to materials distributed through TradingView’s newswire.

The details carry more than one irony. Quinn-Watson became a director of a public company commercializing Eliza roughly ten days before he was named a defendant in the lawsuit that just killed Eliza’s token. And the product that best demonstrates the framework’s future arrives in the same window that its token was declared dead. Strip out the irony, though, and the model is straightforward, and it is the one Walters is now endorsing: sell software and services to customers who pay in dollars, rather than speculation to holders who bet on a ticker.

That is the tokenless thesis in miniature. Revenue from businesses that want the automation, rather than emissions and buybacks funded by a treasury. It looks a lot more like conventional software than like the 2024 agent-token mania. Whether the market for small-business agent automation is large enough to sustain framework development is unproven; the enterprise-automation field is crowded and unforgiving. But it is a business, which is more than the token turned out to be, and it means at least one adopter has a direct financial reason to keep the framework healthy.

The security problem does not die with the token

Whatever maintains Eliza inherits its hardest unsolved problem, and it has nothing to do with tokenomics. Agents that hold wallets and act on natural-language input can be manipulated through that input. The failure mode is not a smart-contract bug; it is that a language model cannot reliably tell an instruction from data.

Eliza specifically was the subject of a widely cited study. Researchers including a Princeton team documented what they called context manipulation, poisoning an agent’s stored memory so that a later, innocent-looking prompt triggers an unauthorized transfer. Their paper, titled around the idea of AI agents with fake memories, introduced a benchmark called CrAIBench and showed that agents built on frameworks like Eliza could be induced to move funds they were never meant to move.

The threat is not theoretical. In May 2026, an attacker drained roughly $216,000 from an AI trading agent by combining an NFT that unlocked transfer permissions with an instruction hidden in Morse code inside a social-media reply, a format that slipped past the model’s plain-text safety filters. The incident is logged by the OECD’s AI incidents monitor, and one security firm called it the first exploit of its kind against a crypto AI agent. The funds were largely recovered; the structural weakness was not.

The proposed fixes all trade away the autonomy that made agents interesting in the first place. Ethereum co-founder Vitalik Buterin argued in an April 2026 essay that autonomous agents should face hard limits, capping the value an agent can move without human sign-off (he floated roughly $100 a day) and treating a human plus the model as a two-of-two approval so neither acts alone. Wallet-level guardrails matter here too, which is part of why permission design in the world of smart-account wallets has become a live topic for agents, not just for people. None of it is solved, and a framework with no foundation to fund audits will be judged on how it handles the next incident.

Eliza against the rest of the field

Eliza is not the only agent framework, and its troubles say little about the others’ code. The comparison worth drawing is between technical footprint and funding model, because that is exactly the axis Eliza just fell off.

The largest agent-framework token by market value is VIRTUAL, from Virtuals Protocol, a no-code launchpad on Base and Solana; it is worth a few hundred million dollars in aggregate, well below its own past peak. Outside crypto, general-purpose orchestration frameworks like LangChain and CrewAI have enormous developer adoption and no token at all, which is precisely the point: the most-used agent tooling in the world was never tokenized, and it is doing fine. Olas, also known as Autonolas, sits in between, an on-chain agent-services network with a live token used for coordination rather than as the product itself. Read across the table and Eliza looks less like an outlier than like the crypto-native branch of a tree whose largest branches never needed a token.

What the SEC did, and did not, do

The regulatory lesson of Eliza is mostly about absence. No SEC action ended this; a private class action did. That reflects where United States crypto policy sat in 2026. In March, the SEC and the CFTC issued a joint interpretation sorting digital assets into categories and naming a set of tokens as digital commodities, but that guidance did not address autonomous AI agents at all. The rulebook was written for spot assets and intermediaries, not for software that signs its own transactions.

There is also a structural gap the Eliza case exposes. An AI agent has no legal personhood and no tax identity, so when one causes a loss, liability falls back on the humans: the deployer, the developer, the user or the model provider, depending on who controlled the failure. That is unsettled law, and it is why misrepresentation claims against identifiable founders, the theory in the Eliza suit, are for now an easier route than novel claims about agent behavior itself.

For anyone tracking how United States oversight actually reaches crypto, the mechanics matter as much as the headlines, in the same way that the formal pipelines behind crypto ETF approvals matter more than the launch-day price action. Eliza’s ending suggests that, for the current cycle, the binding constraint on token projects may be plaintiffs’ lawyers rather than the SEC.

The tokenless turn: what Eliza’s death says about AI-agent crypto

Zoom out and Eliza is a data point in a larger correction. The bet in 2024 was that you could launch a token to fund the development of an open-source AI framework, and that the token’s value would track the framework’s success. Eliza proved the second half false in the most literal way possible: the framework succeeded and the token went to zero anyway.

That does not mean tokens and AI agents are incompatible. It means the naive version, launch a memecoin-adjacent asset, attach a software mission, and hope the two stay correlated, does not hold. The models getting fresh attention now try to bind token value to real usage instead of narrative. Some route protocol revenue into buy-and-burn mechanics; Gensyn’s buy-and-burn design for decentralized machine-learning compute is one closely watched attempt to make a token track actual demand rather than hype. Others, like agent-payment rails, aim to make the token a metered utility. Whether any of them work is an open question, but they are at least responding to the failure Eliza just demonstrated.

The broader AI-agent sector, worth a few billion dollars in aggregate token value and down sharply from its early-2026 highs, is living through the same reckoning over autonomy, security and accountability that this collapse dramatizes. Eliza is not the cause of that reckoning. It is its clearest single illustration: a genuinely useful piece of software whose token was always a separate, and far more fragile, thing.

What to watch next

Four things will tell you whether Eliza’s afterlife is real. The first is Agentic SME: if a public company ships a paying product on the framework this month and finds customers, the tokenless model has a proof point. The second is release cadence: Eliza shipped fast in 2026, and whether that survives the foundation’s dissolution will show up in the commit history within a quarter. The third is fragmentation: watch whether the major forks at teams like Nethermind and Automata converge with Walters’ line of development or drift into incompatible dialects. The fourth is security, because the memory-manipulation problem is the framework’s real overhang.

Walters has already told everyone how he wants it to read. “We’re still gonna be here in 10 years,” he wrote, “pushing at the edge of what agents can do and going where the big corps won’t go.” The token gave him a spectacular, ruinous head start and then took it all back. What is left is the thing he says he wanted to build in the first place: software, given away, that either earns its keep or does not. For once in crypto, the token’s price tells you almost nothing about the answer.

Frequently Asked Questions

Is the Eliza framework shutting down?

No. Shaw Walters declared the ELIZAOS token dead and is winding down its foundation, but the open-source Eliza framework continues. It is MIT-licensed TypeScript on GitHub, still actively developed, and Walters has said he will keep building it with no token attached.

What happened to the ELIZAOS (ai16z) token?

It collapsed. After peaking around a $2.39 billion market cap on January 2, 2025, the token fell more than 99% and now trades near $0.00023 for a market cap of roughly $1.75 million, per CoinGecko. On August 4, 2026, Walters said the foundation had spent its remaining treasury to settle a class action and called the token completely dead.

Why did Shaw Walters kill the Eliza token?

A class action filed by Burwick Law in April 2026 alleged the project was marketed as an autonomous, AI-run fund when humans approved the trades. Walters said the claim was ridiculous but that Eliza Labs lacked the capital to fight it, so it settled by handing over its remaining funds and shut the token down rather than keep supporting it.

Can you still build on ElizaOS?

Yes. The code is open-source and permissively licensed, so developers, companies and community contributors can keep using and extending it. Infrastructure teams including Nethermind and Automata Network already maintain their own forks, and a commercial product called Agentic SME is launching on the framework in August 2026.

Is it safe to run an AI agent built on Eliza?

Agents that hold wallets carry real risks. Researchers have shown Eliza-based agents can be manipulated through poisoned memory or hidden instructions, and a similar attack drained about $216,000 from an AI agent in May 2026. Experts recommend hard limits on how much value an agent can move without human approval.

Reporting by the HOGE Wire ai-crypto desk.