AI and the Future of Social Security — A 2026 Research Survey

What the SSA Trustees actually project

The 2024 Trustees Report (Table V.B1) gives three sets of assumptions: low-cost (favorable), intermediate (central), and high-cost (unfavorable). The intermediate assumption is what most analysts cite. Under intermediate assumptions, the long-range (75-year) actuarial deficit is 3.50% of taxable payroll. The OASI fund is depleted in 2033; combined OASDI in 2035.

The Trustees' sensitivity analysis (Table VI.G2) is the key for AI scenarios. Each 0.5-percentage-point change in long-run real-wage growth changes the 75-year actuarial deficit by about 0.5 percentage points of payroll — roughly a 14% movement in the deficit. The Trustees' central real-wage growth assumption is 1.21% per year long-run. If AI raises that to 1.71%, the actuarial deficit falls to ~3.0%; if to 2.21%, it falls to ~2.5%.

Important caveat: the Trustees model long-run real *wage* growth, not GDP growth. Productivity gains that accrue as capital income or as wages above the $168,600 cap do not move this lever. The Trustees report does not assume any particular AI scenario in its 2024 release.

What CBO, Goldman, and McKinsey project for AI productivity

CBO (August 2024 Long-Term Budget Outlook): assumes generative AI raises long-run labor productivity growth by 0.4 percentage points per year. This is a conservative central estimate intentionally calibrated below the more aggressive private-sector forecasts.

Goldman Sachs (March 2023, Briggs & Kodnani): projects AI could raise US labor productivity growth by ~1.5 percentage points per year over a decade, equivalent to a 7% boost to global GDP. Goldman's note is explicit that this is a 'potentially large' effect, not a forecast.

McKinsey Global Institute (July 2023): projects generative AI could add 0.5–3.4 percentage points per year to US labor productivity, depending on adoption speed. McKinsey's midpoint is around 1.5–2.0 pp.

Stanford HAI 2024 AI Index: documents firm-level productivity gains of 14–55% on specific knowledge-work tasks but emphasizes that the macroeconomic signal is not yet visible. As of Q1 2026, BLS nonfarm-business productivity growth is running at roughly 2.0% YoY — within historical range, not above it.

Why the pass-through to wages is the central uncertainty

Acemoglu & Restrepo's 2022 Econometrica paper 'Tasks, Automation, and the Rise in US Wage Inequality' finds that automation-driven productivity since 1980 explains roughly half to two-thirds of the rise in US wage inequality. Most of the gains accrued to high-skill workers and capital. If AI follows that pattern, GDP grows but the median wage grows less — and the payroll-tax base grows least of all because high-skill income is concentrated above the cap.

The 2024 SSA Trustees Report effectively prices this in by using a conservative real-wage assumption. Goldman Sachs and McKinsey, by contrast, model the productivity boost without specifying how it splits between capital and labor. The implied trust-fund solvency depends almost entirely on this assumption.

Bottom line: if AI productivity is real and accrues to wages, Goldman's 1.5 pp scenario extends OASI solvency from 2033 to roughly 2040, and the long-run actuarial deficit falls from 3.5% to about 1.8% of payroll. If AI productivity is real but accrues to capital, the trust fund hits 2033 on schedule and the actuarial deficit is unchanged.

What the reform plans look like

The Bipartisan Policy Center's 2024 plan would close the entire 75-year shortfall through a combination of: gradually raising the payroll-tax cap to cover 90% of wages (currently ~83%), gradually increasing the Full Retirement Age to 69, modifying the COLA formula (Chained CPI), and modestly raising the payroll tax rate. The BPC plan would solve the funding gap independent of any AI assumption.

The Diamond-Orszag plan (academic baseline) would do similarly through three legs: half of the gap from progressive PIA bend-point reform, a quarter from cap reform, and a quarter from minor tax-rate adjustments.

Bowles-Simpson would solve via a mix of raising the cap, raising the retirement age, and Chained CPI. All major bipartisan plans agree on this basic structure; they disagree mostly on which leg bears more weight.

AI productivity, in the optimistic case, would let any of these plans require smaller adjustments. In the pessimistic case, it would not. Either way, the central uncertainty is political: every plan requires Congress to legislate before 2033, and the political math today is worse than it was in 1977 or 1983.

Frequently asked questions