Anthropic prepares for AI's economic toll with radical tax ideas

Users are no longer just chatting with Claude for quick edits or brainstorming. They are handing over entire workflows and letting the AI work unattended for long stretches.

Anthropic sees this shift as an economic inflection point. On October 14, the AI company teamed up with a bipartisan group of economists to publish nine economic policy proposals aimed at softening the blow of future automation.

The no-regrets moves

Anthropic sorted its ideas into three tiers based on how bad the labor shock gets. The first tier consists of policies that make sense even if AI disruption remains mild.

To keep workers employed, Abigail Ball and Oren Cass of American Compass proposed a Workforce Training Grant that would give employers up to $10,000 per year per trainee, funded by redirecting college subsidies or taxing AI consumption. Mercatus Center's Revana Sharfuddin called for fixing the tax code, which currently lets firms instantly expense AI systems via bonus depreciation while heavily restricting worker retraining write-offs.

Tax expert David Gamage proposed closing corporate tax loopholes and modernizing international apportionment so AI profits from intangibles do not vanish into offshore tax havens.

Anthropic also used the moment to push its own priority: cutting regulatory red tape. Building modern AI centers in the U.S. faces massive bottlenecks, with transmission lines often taking over a decade to clear and grid interconnections taking 4 to 6 years. Anthropic urged reforms to the National Environmental Policy Act to fast-track data centers and power generation.

Token taxes and AI insurance

If job losses pick up speed, the proposals get significantly more aggressive.

UPenn economist Ioana Marinescu suggested an "AI insurance" system based on Trade Adjustment Assistance. Researchers Suchet Mittal and Sam Manning proposed an Automation Adjustment Assistance program kicking off at $700 million annually, with funding scaled up via taxes on high-market-cap AI giants.

Economists Anton Korinek and Lee Lockwood went straight to the technical pipe, pitching taxes on compute and token generation. They argue a token tax makes sense while humans are still the main consumers, but compute taxes might become necessary if AI agents start consuming the bulk of economic resources.

For worst-case scenarios where labor collapses entirely, the group even explored sovereign wealth funds to give every citizen equity in AI revenues.

Why it matters

For workers, the signal is clear: delegating full workflows to agents is no longer theoretical, it is actively showing up in enterprise telemetry. If your job can be packaged into an unattended prompt chain, looking for employer-backed upskilling programs is becoming urgent.

For the tech industry, Anthropic is playing a clever political game. By putting token taxes and corporate loop-closing on the table, it positions itself as the grown-up in the room willing to tax its own outputs.

Yet, we think there is obvious self-interest mixed in. Anthropic paired humanitarian safety nets directly with permitting reform, effectively telling governments: let us build massive power plants and data centers faster, and we will help you figure out how to tax the fallout later. Whether a token tax would actually protect workers or simply push compute clusters into tax havens remains an unanswered question.