The number is cold. 45% of Americans aged 18 to 34 expect AI to hurt their careers. Only 10% see it as a helper. This is not a survey. It is a signal. The signal is not about fear of machines. It is about the failure of the current economic consensus layer. Andrew Yang, the former presidential candidate, is back on CNBC’s Power Lunch. He is not talking about universal basic income this time. He is pushing for an AI tax. The logic is simple: firms that replace human workers with AI models should pay a tax on the revenue generated by those models. The tax should replace payroll taxes. The government should stop taxing labor. This is a proposal that sounds like a technical fix. But as a Layer2 Research Lead who has spent years auditing smart contracts and untangling protocol failures, I see a different problem. The proposal is not about economics. It is about verifiability. How do you tax a black box? The code does not lie, but the auditor must dig. And the first thing I dig into is the assumption that the government can measure AI revenue at all.
Tracing the gas trails back to the root cause: the root cause is not automation. It is the inability of the current fiscal system to track value creation in a machine-driven economy. Yang’s argument is straightforward. Companies skip payroll taxes and healthcare costs by choosing AI over new hires. So the government should tax the AI instead. He points to Anthropic CEO Dario Amodei, who floated a 3% AI revenue tax in 2025. The idea is to apply the levy each time a model generates revenue. Bridgewater Associates executives Greg Jensen and Nir Bar Dea echoed this in a New York Times opinion piece, estimating AI could displace 18% of current US jobs within five years. They proposed an AI token tax. The pattern is clear: the political and financial establishment sees a need to capture value from AI. But the mechanism is undefined.
Context: The current proposal is a political artifact, not a technical specification. Yang’s background is instructive. He built his 2020 campaign on automation warnings and the Freedom Dividend. He also backed cryptocurrency adoption and clearer digital asset rules. This is not a coincidence. The same forces that drive crypto adoption—distrust of centralized intermediaries, desire for transparent value transfer—are at play in the AI tax debate. But the AI tax proposal, as currently framed, relies on the same legacy infrastructure that crypto was designed to replace. The government would need to track AI revenue, enforce compliance, and redistribute funds. This is a software problem. And software problems require cryptographic proofs, not political will.
Shifting the consensus layer, one block at a time: the AI tax proposal is a consensus failure. It assumes that the government can observe and verify the revenue generated by an AI model. But AI models are not like employees. They do not have W-2 forms. They do not have time sheets. Their revenue is often embedded in complex pipelines—automated trading, customer service chatbots, generative content. The revenue is not attributed to a single model. It is aggregated across multiple models, data sources, and human oversight. The tax authority would need to parse this complexity. Based on my experience reverse-engineering the Terra-Luna collapse, I can tell you that parsing complex systems is not a matter of good intentions. It is a matter of correct state machines. The Terra protocol had a mathematical instability. The AI tax proposal has a verifiability gap.
Core: Let me break this down at the protocol level. The AI tax proposal has three components: identification, measurement, and enforcement. Identification: which AI models are subject to the tax? Is it generative models, decision models, or all models? The boundary is fuzzy. A model that recommends products is not the same as a model that writes code. But the tax code would need to define a taxonomy. This is like defining a smart contract interface without a standard. In blockchain, we have ERC-20, ERC-721, and so on. For AI, there is no equivalent. The government would need to audit every company’s internal AI stack. This is impractical.
Measurement: how do you measure the revenue generated by an AI model? The model does not have a bank account. The revenue is collected by the company. The company can attribute revenue to the model, but that attribution is opaque. In my work auditing the Parity Multisig, I learned that the simplest vulnerabilities are often in the accounting logic. The Parity wallet had a kill function that allowed any user to drain funds. The AI tax proposal has a similar vulnerability: the ability to misattribute revenue. A company could claim that the AI model contributed zero revenue, while the human employees contributed everything. The tax authority would need to audit the company’s internal metrics. This is not a new problem. It is the same problem that plagues transfer pricing in multinational corporations. The solution is not a tax rate. It is a cryptographic proof of work.
Enforcement: how do you collect the tax? Yang proposes sending the revenue directly to workers as checks. But the distribution mechanism is untested. Retraining programs have failed, as Yang himself notes. He points to coal miners and warehouse staff. The failure is not due to lack of funding. It is due to lack of alignment. The workers are not being retrained for jobs that exist. The AI tax would create a new class of transfer payments. But the infrastructure for those payments is the same outdated banking system. The tax revenue would flow through government agencies, which are slow and prone to error. This is where blockchain could help. A smart contract could automatically distribute tax revenue to verified workers based on on-chain identity. But that requires a decentralized identity system, which is still in its infancy.
Contrarian: The blind spot in the AI tax proposal is not the tax itself. It is the assumption that the government can execute it efficiently. Yang’s argument is that the government should tax AI instead of labor. But the government is a centralized entity with a poor track record of technology adoption. The IRS uses 1960s-era mainframes. The Social Security Administration is still processing paper forms. The idea that the same government can implement a sophisticated AI tax system is naive. This is not a political statement. It is a technical observation. In my work on the StarkNet recursive proofs investigation, I saw how complex cryptographic systems require careful state management. The government does not have the cryptographic infrastructure to manage an AI tax.
Furthermore, the AI tax proposal could be gamed. Large companies with legal teams can structure their operations to minimize tax liability. They can create shell companies in jurisdictions with no AI tax. They can open-source their models and claim that the revenue is from services, not the model itself. The tax would become a burden on small startups, not big tech. This is the same pattern we see in crypto regulation: compliance costs are passed to honest users. The code does not lie, but the tax code does. The proposal is a political gesture, not a technical solution.
Another blind spot: the AI tax does not address the root cause of job displacement. The root cause is not the cost of labor. It is the structure of the economy. We have a system that rewards automation over human labor because labor has a tax burden. Yang’s proposal flips that: it taxes AI instead. But the underlying incentive remains the same. Companies will still automate because AI is more efficient. The tax just adds a cost. The question is whether the tax is high enough to change behavior. If it is too high, it stifles innovation. If it is too low, it is irrelevant. The optimal tax rate is a political calculation, not a technical one.
Takeaway: The future of work is not about taxing AI. It is about redesigning the incentive structures. The blockchain community has a unique opportunity here. We can build decentralized autonomous organizations that distribute AI-generated value equitably. We can create transparent, verifiable systems for tracking value creation and distribution. The AI tax proposal is a step in the right direction, but it is built on legacy infrastructure. The real solution is a new consensus layer—one that can handle the complexity of a machine-driven economy. Shifting the consensus layer, one block at a time. The code does not lie, but the auditor must dig. And the auditor is looking at the government’s proposal with the same skepticism I applied to the Terra-Luna peg mechanism. The math is not adding up. The tax is a band-aid. The wound is deeper. The wound is the lack of a verifiable economic identity for AI.
Based on my experience designing an AI-agent on-chain identity framework, I know that the first step is to create a decentralized identity for AI models. Each model should have a unique identifier, a public key, and a record of its computational work. This record can be used to verify revenue attribution. Zero-knowledge proofs can ensure privacy while proving accuracy. The government could then audit the aggregated proofs, not the individual transactions. This is the same approach I used in my StarkNet research: recursive proofs that compress multiple verifications into one. The AI tax could be implemented as a smart contract that accepts proof of revenue and issues a tax receipt. The distribution to workers could be automated via a DAO. This is not science fiction. It is engineering.
But the political will is missing. The government is not ready to adopt cryptographic infrastructure. The AI tax proposal is a political signal, not a technical blueprint. As a researcher, I see the gaps. As a citizen, I see the urgency. The 45% of young Americans who expect AI to hurt their careers are not wrong. They are early. The question is whether we will build a system that captures the value of AI for the many, or for the few. The code does not lie. But the choices we make about the code do. The AI tax proposal is a start. But it is not the end. The end is a new economic consensus layer, built on transparency, verifiability, and fairness. And that layer will not be built by politicians. It will be built by engineers.