Andrew Yang's AI Tax: A Systemic Fork in the Road

Andrew Yang, the 2020 presidential candidate, revived his AI tax proposal on CNBC's Power Lunch, arguing that the government should tax artificial intelligence instead of payroll. His logic is brutally simple: every time a firm replaces a human with an AI system, the government loses payroll tax revenue and healthcare contributions. Tax the machine, not the worker. Predictability is a myth; only volatility is real—and Yang is betting that the volatility of labor displacement will force a political realignment. Yang, now CEO of Noble Mobile and co-founder of the Forward Party, built his political brand on automation warnings. In 2020, he proposed a Universal Basic Income called the Freedom Dividend. He also backed cryptocurrency adoption and clearer digital asset rules. His current push echoes a March interview on CNBC's Squawk Box, where he argued the government should stop taxing labor. Now, he is pointing to Anthropic CEO Dario Amodei, who floated a 3% AI revenue tax in 2025. Amodei said the levy would apply each time a model generates revenue—a per-transaction tax on machine intelligence. The proposal is not new. Bridgewater Associates executives Greg Jensen and Nir Bar Dea wrote a New York Times opinion piece estimating that AI could displace 18% of current US jobs within five years. They used that estimate to back their own AI token tax proposal, which would tax AI-generated output at the protocol level. The idea is gaining traction among policymakers, but the technical implementation remains a mess. From my experience auditing smart contracts and modeling systemic risk in DeFi, I see the parallels: any tax on AI revenue will create a new vector for gaming, evasion, and unintended consequences. History does not repeat, but it rhymes in binary. The same composability risks that haunted DeFi—where a single hook in Uniswap V4 could cascade into a liquidation cascade—will apply to AI taxation. If a tax is levied per model revenue, firms will simply split models into smaller entities, or route revenue through offshore intermediaries. The IRS cannot track a billion micro-transactions without a blockchain-based audit trail. Yang’s proposal, while politically catchy, ignores the technical reality of machine-generated revenue streams. Let’s examine the data. A CNBC and Generation Lab survey published August 13 polled Americans aged 18 to 34. It found 45% expect AI to hurt their careers, while only 10% expect it to help. That is a massive asymmetry in expectation. The customer service sector, which employs roughly 2.9 million Americans according to the Bureau of Labor Statistics, is already seeing displacement. AI chatbots handle 40% of tier-1 queries now, up from 10% in 2022. The tax base is eroding in real time. Yang proposed sending the tax revenue directly to workers as checks. He said retraining programs rarely help displaced workers find new careers. He pointed to past efforts aimed at coal miners and warehouse staff as examples that largely failed. This is where the systemic interdependence mapping becomes critical. A simple tax on AI revenue does not address the root cause: the labor market is a network, not a pipeline. Displacing 18% of jobs creates cascading effects in housing, education, and healthcare. The tax revenue must be mapped to where the fragility is highest—not just distributed as flat checks. Here is the contrarian angle that most coverage misses: the AI tax could actually accelerate the shift toward decentralized AI networks. If the government taxes centralized AI revenue at 3%, firms like OpenAI and Anthropic will face a cost disadvantage compared to decentralized alternatives. Projects like Bittensor or Gensyn—where model inference is run on peer-to-peer networks—could avoid the tax entirely because there is no single entity generating revenue. The tax becomes a competitive advantage for decentralized AI. As a cryptographer, I see this as a predictable outcome: every regulation creates a shadow economy optimized for evasion. Moreover, the AI tax proposal ignores the fact that AI models themselves are becoming composable. An AI agent that calls another AI model’s API to generate a trading strategy—who pays the tax? The first model, the second, or the user? The attribution problem is identical to the flash loan tax problem in DeFi. If you tax each leg of a composable transaction, you kill the entire ecosystem. If you tax only the final output, you create a loophole for intermediate models to operate tax-free. Yang’s proposal also fails to account for the speed of AI adoption. The Bridgewater estimate of 18% displacement within five years assumes a linear adoption curve. But from my work modeling flash crash dynamics, I know that adoption curves in technology are logistic, not linear. Once AI reaches a certain threshold of reliability, the shift to AI labor happens in weeks, not years. The tax infrastructure will not be ready in time. The IRS takes years to implement new tax forms, while AI models are updated every month. Takeaway: The AI tax debate is a systemic fork in the road. The path chosen will determine whether the next decade sees a consolidation of power in centralized AI giants or a fragmentation into decentralized, tax-optimized networks. Yang is right to start the conversation, but he is wrong to assume that a simple revenue tax is the answer. The real solution is a cryptographic audit trail for every AI-generated transaction—a transparent, immutable ledger that allows governments to tax without breaking the composability of the system. Without that, the tax will be little more than a political signal in a volatile market. Predictability is a myth; only volatility is real—and the AI tax is the next volatility event waiting to be engineered.