Step 1: Choose a meme to share
AI Legal Enforcement creates an FBI division deploying government 'White Hat' AI agents to investigate AI-driven federal crimes (built on existing FBI, CFAA, and securities-law authority) and funds it via a $1-per-million-output-token excise on firms exceeding 1 billion tokens/year. The tax is vastly larger than the division needs and than the marginal cost of cheap models, so it functions as a broad, distorting AI revenue tax that reduces federal debt by an estimated ~$175 billion over ten years (wide uncertainty) while raising surveillance, First Amendment, and avoidance concerns. Incidence falls progressively on shareholders and AI-service users, imposing small costs on lower-income households and much larger costs on the top 1% and 0.1%.
This proposal pairs a novel federal law-enforcement structure with a novel excise tax. It would (1) create an AI-enforcement division inside the FBI staffed by AI researchers deploying 'White Hat' agentic systems to investigate AI-driven federal crimes and build cases for DOJ prosecution, and (2) fund that division through a per-token excise tax of $1 per million output tokens levied on any company generating more than 1 billion tokens per year. The reform is NOT revenue-neutral: as drafted, the tax rate is fixed and would raise far more revenue than an FBI division could plausibly spend, so the excess flows to the general Treasury and reduces the deficit.
The Federal Bureau of Investigation is established within the Department of Justice under Chapter 33 of Title 28. The Attorney General's authority to appoint officials to detect and prosecute crimes against the United States flows from 28 U.S.C. Section 533, delegated in practice to the FBI Director. The substantive offenses this division would pursue already exist: computer intrusion under the Computer Fraud and Abuse Act (18 U.S.C. Section 1030), securities fraud and market manipulation under Section 10(b) of the Securities Exchange Act (15 U.S.C. Section 78j), and insider-trading civil penalties under 15 U.S.C. Section 78u-1. Corporate criminal liability already attaches to companies whose agents commit federal crimes under respondeat superior principles. On the tax side, federal excise taxes are grouped in Subtitle D of the Internal Revenue Code and use the operative 'There is hereby imposed a tax' formula (e.g., the petroleum tax at 26 U.S.C. Section 4611). No existing statute imposes a tax on AI token generation or defines an 'AI agent' as a criminal actor.
The bill would insert a new division into Chapter 33 of Title 28, inheriting existing FBI investigative authority rather than creating standalone powers. It would authorize the deployment of government-operated AI systems as investigative tools using 'all standard approaches of law enforcement,' and it would create a new excise tax in Subtitle D of the IRC. It also contemplates that Congress may enact new AI-specific crimes. Importantly, the bill assigns criminal culpability to the individuals or corporations running the offending AI, not the AI itself, which is legally coherent since software cannot bear criminal intent (mens rea).
For: As agentic workloads scale, autonomous systems can plausibly be used to commit hacking, fraud, and market manipulation at machine speed, outpacing conventional investigators. A specialized, technically sophisticated unit could close a genuine enforcement gap, and an industry-funded model avoids burdening general taxpayers. Sizing enforcement funding to token volume creates a rough proxy for the scale of AI activity being policed.
Against: The tax is dramatically oversized relative to its stated purpose (an FBI division would cost, at most, low single-digit billions annually, while the tax would raise far more), making the 'funding' rationale largely pretextual and turning the measure into a broad AI revenue tax. The flat $1/million-token rate is roughly 25x the marginal price of the cheapest 2026 models (about $0.04/million tokens) and, with token prices falling ~80% year-over-year, the wedge grows over time. This would crush low-margin and open-source inference, distort model design toward token efficiency, and push generation offshore or into non-corporate structures to escape the corporate threshold. Government-operated 'White Hat' agents conducting investigations raise serious surveillance, entrapment, and civil-liberties concerns.
The enforcement provisions are largely constitutional in structure because they build on existing FBI/DOJ authority and existing crimes; the principal Fourth Amendment concerns arise from how agentic surveillance tools gather evidence, which would require warrants and would generate substantial suppression litigation. Due-process and vagueness challenges are likely for any new AI-specific crimes and for attributing an AI's conduct to a corporate operator. The tax is defensible as an excise under Congress's broad Article I taxing power; excise taxes need only be geographically uniform, and a per-token levy satisfies uniformity. First Amendment questions could arise because taxing 'output tokens' is effectively taxing generated expression, inviting scrutiny akin to differential taxation of speech/press, though a facially content-neutral, activity-based excise would likely survive.
Best estimate is that this proposal reduces federal debt over ten years, on the order of $175 billion, though the confidence interval is very wide. Sizing the tax base: OpenAI alone processed about 15 billion tokens per minute in early 2026 (roughly 7.9 quadrillion combined tokens per year), and the global API market runs near 50 trillion tokens per day. Assuming output is roughly one-third of combined volume, U.S. large-company output tokens in 2026 are on the order of 10-15 quadrillion per year. At $1 per million output tokens (equivalent to $1 million per trillion, or $1 billion per quadrillion), the static base implies roughly $10-15 billion per year initially. Explosive volume growth would push gross receipts far higher over the decade, but three forces sharply offset this: (1) strong behavioral avoidance, since the tax vastly exceeds marginal cost for cheap/open-source inference and would drive volume offshore, toward efficiency, and below thresholds; (2) corporate-income-tax feedback, as the excise is deductible and reduces income-tax receipts by roughly 21 cents per dollar; and (3) new spending for the FBI division (plausibly $0.5-1.5 billion per year, or roughly $8-12 billion over ten years). Netting explosive-but-suppressed gross receipts against avoidance, income-tax offset, and program spending yields an estimated net debt reduction of about $175 billion over ten years. Given the extreme sensitivity to token-volume growth and avoidance elasticity, plausible outcomes range from under $75 billion to over $400 billion in debt reduction.
The statutory incidence is on large AI companies, but economic incidence splits between shareholders (through lower after-tax profits) and consumers/business users (through higher AI service prices). Equity ownership is heavily concentrated at the top of the distribution, so the shareholder portion is steeply progressive. The consumer-pass-through portion tracks AI spending, which currently skews toward higher-income households and enterprises but is broadening. Net household impacts are modeled as costs (negative), reflecting higher prices for AI-enabled goods/services and reduced investment returns, with the diffuse benefit of deficit reduction not attributed to specific households. The burden is small for lower quintiles and rises sharply for the top 1% and top 0.1%, making the measure progressive overall. Limitations: household-level incidence of a business-input tax is inherently approximate, pass-through rates are uncertain, and rapid market growth plus avoidance could shift these figures materially.
What this means: This shows how the proposal would raise or lower the nation's debt. It also shows the change on a per household basis, assuming the debt burden was evenly distributed.
This proposal will decrease the USA's debt by $175 billion over 10 years. This is equivalent to decreasing the debt by $1,336 per American household.
What this means: The table shows the proposal's impact on household income by income class. It shows which groups, rich or poor, benefit or bear costs.
| Household Income (per Year) | Annual Impact |
|---|---|
|
<$30K Lower class (Bottom 20%) |
-$5 (-0.0%) |
|
$31K-$59K Lower-middle class (20-40%) |
-$12 (-0.0%) |
|
$60K-$95K Middle class (40-60%) |
-$25 (-0.0%) |
|
$96K-$160K Upper-middle class (60-80%) |
-$55 (-0.1%) |
|
>$160K Upper class (Top 20%) |
-$180 (-0.1%) |
|
>$590K Top 1% |
-$1,200 (-0.1%) |
|
>$2.4M Top 0.1% |
-$8,000 (-0.1%) |
(For econ/math nerds: the Gini index decreases 0.0% from 0.5285 to 0.5284)
Step 1: Choose a meme to share
No comments yet. Be the first to share your thoughts!