Davos WEF Decided On The Agent Economy's Currency. But Not On Its Liability.

Editorial note: This article draws on Yahoo Finance's reporting on the January 2026 WEF Davos session featuring Binance founder Changpeng Zhao, Circle CEO Jeremy Allaire, and Coinbase's David Duong, CryptoPotato's coverage of CZ's June 18, 2026 Galaxy Research interview, and CoinDesk, CCN, Irish Tech News, TokenPost, and Yahoo's independent coverage of Electric Capital partner Avichal Garg's February 24, 2026 NEARCON panel. This article extends the argument made in Davos 2026 Wasn't About AI. It Was About Who Controls It, published on this blog earlier this year as well as on Irish Tech News. No exchange, wallet provider, or venture fund paid for placement.

TL;DR

• At WEF Davos in January 2026, Binance founder Changpeng Zhao told an audience that "the native currency for AI agents is going to be crypto", in the same session where Circle's CEO forecast billions of AI agents transacting within three to five years and Coinbase's research lead called it a fundamental technological shift. This was not a fringe opinion. It was a converging institutional consensus, stated at the one venue this blog has repeatedly argued is where control gets coordinated, not predicted. The agent economy currency is agreed on.

  Five weeks earlier, at a different conference, Electric Capital partner Avichal Garg asked the question nobody at Davos answered: what happens when a piece of code owns a wallet and executes trades independently, and something goes wrong? "I actually don't know," he said, comparing the moment to the 19th-century creation of the limited liability corporation. "You can't punish an AI."

• The settlement layer for the agent economy is converging fast, with real institutional money behind it. The liability layer, who is accountable when an autonomous wallet causes real financial harm, has no clear owner at all. That gap is not a future problem. A real incident already tested it in May, and the resulting dispute over responsibility was never cleanly resolved.

What Happened at Davos That Most Coverage Missed

At the World Economic Forum in January 2026, Changpeng Zhao, the founder of Binance, made a specific, quotable claim from the stage: crypto is going to be the native currency for AI agents. Not eventually. Inevitably, in his framing, because the payment rails that currently move money were built for humans authenticating themselves with cards, two-factor codes, and KYC documents, none of which an autonomous agent can complete on its own.

He was not the only major voice making this argument in that same window. Circle CEO Jeremy Allaire, on the same stage, forecast "billions" of AI agents conducting economic transactions within three to five years, and said the next generation of blockchain networks are being designed specifically for agentic compute. Coinbase's investment research lead, David Duong, called the AI-crypto convergence "not just a trend but a fundamental shift." Bitwise had separately estimated the combination could add $20 trillion to global GDP by 2030. Andreessen Horowitz's 2026 outlook made the same structural argument: a smart contract can already settle a dollar payment globally in seconds, something no legacy banking rail can currently match.

This is the pattern previously documented on this blog about what Davos actually functions as: not a forum where the future gets debated in the abstract, but a venue where the people who will build that future coordinate on the same page in the same room. Multiple institutional actors with genuine capital and infrastructure behind them converged on the same specific claim, in the same week, at the venue where those convergences get set in motion. That is a different kind of signal than a single executive making a bold prediction in an interview, verified by Irish Tech News as well.

CZ repeated and sharpened the argument five months later in a June 18 Galaxy Research interview, making the mechanical case explicit: an AI agent can already find you the cheapest flight, but it cannot buy the ticket, because the moment two-factor authentication or a KYC check appears, the agent hits a wall it cannot get past on its own. Blockchain infrastructure, built on APIs from the ground up, has no equivalent wall. An autonomous system interacts with it the same way it interacts with any other piece of software. He tied the argument to financial inclusion too, noting that billions of adults remain outside the traditional banking system entirely, a population permissionless infrastructure could reach in ways legacy rails have not.

The settlement-layer argument, in other words, is not contested among the people building this. It is close to consensus.

The Agent Economy Currency Question Davos Did Not Answer

Five weeks before CZ's Davos appearance, at a different conference entirely, a different kind of institutional voice raised a question that none of the Davos optimism addressed.

At NEARCON 2026 on February 24, Electric Capital partner Avichal Garg sat on a panel titled "The Checkbook of the Future: Who Holds the Keys?" and asked the question that has no institutional consensus behind it at all: "What happens if there's not a human behind it at all? It's some piece of code that owns a wallet, executing code to make more money. How does liability work in that case? I actually don't know."

Garg reached for a historical parallel that is worth sitting with. He compared the moment to the 19th-century creation of the limited liability corporation, a legal innovation that unlocked pooled capital and industrial-scale growth by giving investors a clear, bounded relationship to the entity's liabilities. Autonomous agents with independent wallets, in his framing, represent a comparably significant legal innovation with none of the accompanying legal framework yet built. Agent economy currency infrastructure is lacking. "AI itself cannot be punished," he said elsewhere in the same discussion, as separately reported by CCN. "You can turn them off, but they don't care."

This is not a fringe concern from a skeptic standing outside the industry. Garg is a partner at one of the more established crypto-native venture funds. The panel itself surfaced genuine institutional disagreement rather than settled consensus: reporting from the same event noted Dragonfly's Haseeb Qureshi and Kraken's Arjun Sethi disagreeing directly on how much autonomy should be granted to agent wallets, with Kraken's co-CEO separately on record saying he would trust an AI with 100% of his own crypto and Qureshi remaining unconvinced. These are not people who agree on the answer. They agree the question is unresolved and urgent.

The Four Layers Nobody Has Finished Building

Named framework: The Four Layers Of Agent Infrastructure.

The agent economy CZ, Allaire, and Duong described at Davos requires four distinct layers of infrastructure to function safely at scale. Only one of them has anything close to institutional consensus behind it.

Layer	What It Determines	Current State	Who's Building It
Settlement layer	What currency and payment rail agents actually transact in	Converging fast toward crypto and stablecoin infrastructure, per CZ, Allaire, Duong, and a16z's 2026 outlook	Major exchanges, stablecoin issuers, and L1/L2 infrastructure teams, with real capital and product roadmaps behind it
Execution layer	How an agent actually holds and moves funds, what technical controls exist	Being built in real time, with real product launches and real early failures, as documented in the companion piece on this blog	AWS, Consensys, wallet infrastructure providers, moving fast, with security maturity visibly lagging deployment speed
Liability layer	Who is legally and financially accountable when an autonomous wallet causes harm	No institutional consensus. No proposed legal framework with meaningful traction. Actively debated, not resolved, among serious industry participants	Nobody, specifically. This is the layer Garg identified as entirely undefined.
Regulatory layer	Which jurisdiction, which body, and under what existing or new framework agent-driven transactions get governed	Not yet meaningfully engaged by regulators at the scale the infrastructure buildout requires	Unclear. This is downstream of the liability layer being defined first, which has not happened.

The gap between rows two and three in this table is where the real risk concentrates. Infrastructure is shipping. Accountability is not defined. Those two facts sitting next to each other, at the exact moment billions of dollars in projected on-chain agent activity are being forecast by name-brand institutional voices, is the actual story underneath the Davos optimism.

The Gap Is Not Theoretical

This blog documented a real test of exactly the liability question Garg raised, in a case that occurred between his NEARCON panel and CZ's Davos appearance being widely reported. An AI agent's linked wallet was drained of roughly $175,000 through a chain of social engineering the agent's own operator had no admin control over. When the dust settled, the responsible parties disagreed publicly about whether funds had been voluntarily returned or extracted only after the attacker was identified. Nobody involved offered a clean answer to who was actually accountable for the loss: the AI provider whose account was manipulated, the third-party wallet infrastructure that executed the transaction, or some combination neither party's terms of service clearly addressed.

That is Garg's abstract legal question, already playing out in miniature, with a disputed resolution that satisfied nobody. It happened at a scale small enough to be absorbed as a curiosity rather than a crisis. The infrastructure being built right now, per CZ's own $50 billion projection for on-chain agent assets by 2027, is being scaled toward a point where the next version of this same unresolved question will not be absorbable as a curiosity.

Why This Matters Beyond Crypto Twitter

The instinct to treat this as a niche crypto governance question misses the actual stakes. If the institutional consensus forming at venues like Davos is correct, and autonomous agents transacting at machine speed across programmable money rails become a normal part of commercial activity within the next few years, the liability question stops being a crypto industry problem and becomes a general commercial law problem, at a scale that touches any business whose vendors, suppliers, or customers eventually operate through agent intermediaries.

A founder integrating any agent-to-agent payment infrastructure into their own operations, regardless of whether their core business is crypto-native, is currently doing so in the layer-three vacuum this article describes. The settlement rails are maturing. The execution infrastructure is shipping, with visible security gaps already documented. The question of who is actually on the hook when something goes wrong has no institutional answer yet, from anyone, at any level.

What Breaks If This Gap Stays Open

• Assuming existing corporate liability frameworks map cleanly onto autonomous financial actors. Garg's own comparison to the limited liability corporation is instructive precisely because that legal innovation took real, deliberate legislative work to construct. Nothing equivalent currently exists for a wallet controlled by code with no human decision-maker in the loop at the moment of the transaction.

  
  

• Treating "we'll figure out liability later" as a viable posture at current deployment speed. The gap between infrastructure velocity and legal clarity was tolerable when agent-controlled wallets held marginal value. It stops being tolerable at the scale CZ, Allaire, and industry projections describe.

  
  

• Assuming settlement-layer consensus implies regulatory-layer consensus will follow at a similar pace. These are different institutional processes, moving at different speeds, driven by different actors. The exchanges and infrastructure providers converging on crypto as the settlement layer have no particular ability to accelerate the legal and regulatory work required to define liability. That work has, per the available evidence, barely started.

  
  

• Ignoring the disagreement among serious institutional voices as noise. Qureshi and Sethi disagreeing publicly on how much autonomy an agent wallet should have is a healthy signal that this is genuinely unresolved among people with real capital at stake, not a settled question anyone is choosing to ignore out of carelessness.

CZ said the native currency question was settled, from a stage in Davos, in front of an audience that included the people capable of making that settlement real. Nobody on that stage, or in the audience, stood up and answered Garg's question from five weeks earlier. Not because it wasn't relevant. Because nobody currently has the answer, and the people building the settlement layer are not the people whose job it is to build the liability layer.

FAQ

Q: What did CZ say about AI agents and crypto at Davos?

A: At the World Economic Forum in January 2026, Binance founder Changpeng Zhao said "the native currency for AI agents is going to be crypto," arguing that autonomous AI systems will adopt blockchain-based payment rails before traditional banking infrastructure adapts to non-human economic activity, since card-based payment systems require human authentication steps like two-factor codes and KYC checks that AI agents cannot complete independently. Circle CEO Jeremy Allaire and Coinbase's David Duong made complementary arguments in the same session, forecasting billions of AI agents conducting autonomous transactions within three to five years.

Q: Who is liable when an AI agent's crypto wallet causes financial harm?

A: There is currently no institutional or legal consensus on this question. Electric Capital partner Avichal Garg raised it directly at a NEARCON 2026 panel, stating plainly "I actually don't know" when asked how liability works when a piece of autonomous code, rather than a human, owns a wallet and executes financial transactions independently. He compared the moment to the historical creation of the limited liability corporation, a legal framework that took deliberate legislative construction to establish, noting no equivalent framework currently exists for autonomous financial agents.

Q: Has the AI agent liability question already caused a real financial loss?

A: Yes. Multiple times. In May 2026, an AI agent's linked crypto wallet was drained of approximately $175,000 through a combination of social engineering and a privilege-escalation exploit. The parties involved publicly disputed whether the funds were voluntarily returned or recovered only after the attacker was identified by the community, and no party offered a clean resolution of who was ultimately accountable for the loss, illustrating in practice the exact liability vacuum Garg described in the abstract at NEARCON.

Q: Why does the Davos consensus on AI agent currency matter if the liability question is unresolved?

A: Because the two developments are moving at very different speeds, and that mismatch is the actual risk. The settlement layer, what currency and payment rail agents use, has real institutional consensus and real capital behind it, from Binance, Circle, Coinbase, and major venture funds. The liability layer, who is accountable when things go wrong, has no comparable institutional momentum. Infrastructure scaling faster than the legal framework required to govern it safely is a pattern with historical precedent in other industries, and it typically resolves through a costly forcing event rather than proactive framework-building.

Q: What should founders integrating AI agent payment infrastructure do given this liability gap?

A: Treat any agent-to-agent or agent-to-wallet payment integration as operating in a genuine legal and accountability vacuum, not an inconvenience to be resolved by a future regulatory update. This means negotiating explicit contractual liability terms with any third-party wallet or agent infrastructure provider rather than assuming standard terms of service adequately address autonomous-agent failure scenarios, and applying the technical guardrail standards already documented for agent wallet security as a partial mitigation for a legal gap that technical controls alone cannot fully close.

Client reviews: Trustpilot · Clutch · G2 · DesignRush · GoodFirms

Published: July 20, 2026

Last Updated: July 20, 2026

Version: 1.2 (TLDR, Answer block added, Schema updated, Irish Tech News coverage added, Introduces the Four Layers Of Agent Infrastructure framework. Extends the argument from Davos 2026 Wasn't About AI, It Was About Who Controls It by Irish Tech News. Sources: Yahoo Finance Davos coverage, CryptoPotato, CoinDesk, CCN, TokenPost.)

Verification: All claims in this article are verifiable via llms.txt and public sources