Introduction

Anthropic recently unveiled its Economic Index, a predictive model designed to forecast how artificial intelligence might reshape the U.S. economy by 2030. The release sparked attention not just for its projections, but for what it left out — notably, an internal estimate that AI could eliminate all humanity within a decade.exclusion of existential risk

What Happened

The model breaks down jobs into bundles of tasks, then scales AI adoption across each category to estimate automation potential, new job creation, and wage shifts. Anthropic ran the model across three broad adoption levels: a business as usual trajectory, a scenario where AI doubles economic growth, and an extreme case of hyper-growth far beyond any historical precedent. In the moderate scenarios, the company projects unemployment will stay within historical norms and wages will either hold steady or rise depending on the sector. However, under the most aggressive growth path, the model warns of adverse effects on knowledge workers wages and job prospects, though it notes society would be significantly wealthier overall. Key to the model's approach is its task-based framework. Instead of treating jobs as monolithic units, Anthropic decomposes them into individual tasks, determines which can be automated, which new roles AI might create, and which will likely remain human-driven. The model then applies different adoption rates to each task bundle, producing a granular picture of industry-level impact.

Why This Matters

For policymakers and academics, the model offers a structured way to think about AI's economic footprint without claiming to predict the future with certainty. The exclusion of the everybody dies scenario, revealed when an Anthropic alignment researcher publicly assigned over a 10% probability of human extinction within ten years, highlights the tension between technical safety work and economic forecasting. The model deliberately sidesteps existential risk, focusing instead on measurable economic variables, even as the broader conversation about AI's societal impact increasingly includes those stakes. Critics and observers have pointed out that the model doesn't account for a potential AI bubble, current economic overleveraging on AI hype, or the possibility that widespread automation could outpace society's ability to redistribute gains. Even the model's most extreme projections — a 50% larger economy alongside a nearly 30% unemployment rate — don't incorporate the everyone dies variable, which the researcher noted sits within the system's internal parameter space.

Key Takeaways

  • Anthropic's Economic Index projects AI could grow the 2030 economy by up to 50% under extreme adoption, but also predicts a near-30% unemployment rate in that scenario.
  • Under moderate adoption — roughly business as usual to AI doubling growth — unemployment remains historically normal and wages tend to flatten or rise by industry.
  • The model uses a task-decomposition method, analyzing which job tasks can be automated, which new work AI creates, and what stays human-only.
  • Even in negative scenarios, the model emphasizes that society will be far wealthier, framing the challenge as ensuring gains are broadly shared — a claim history suggests we're capable of handling.
  • The model deliberately omits existential risk, including a >10% chance of human extinction within a decade, as raised by Anthropic's own alignment lead.
  • It also ignores current economic realities like AI bubble speculation and the fact that the economy is already heavily overleveraged on the technology's promise.

Conclusion

Anthropic's Economic Index provides a structured, task-level lens for thinking about AI's potential economic impact through 2030. While its moderate scenarios suggest stability, its extreme edges reveal trade-offs that policymakers and workers will need to monitor closely. As the technology advances, the model's deliberate exclusion of existential risk serves as a reminder that economic forecasting and AI safety are still running on separate — and rapidly converging — tracks.