Paralegals and Legal Assistants

BLS Occupational Outlook 2024-34

→ 2034

BLS Employment Projections — industry-occupation matrix + labor productivity assumptions. The current published outlook for 23-2011 (2024-34 cycle): "little or no change" — essentially flat employment, with net openings of 376,800 projected vs. 376,200 in 2024 (+600, rounding to 0%). Despite flat net growth, 39,300 annual openings are projected from turnover and retirement. The BLS explicitly cites AI as the mechanism: "advances in technology, including artificial intelligence, are expected to make paralegals and legal assistants more efficient at tasks such as conducting research and preparing documents, which may reduce demand for these workers." This is the first BLS projection cycle to explicitly name AI as a constraint on paralegal employment growth — a sharp departure from every prior decade, in which this occupation was consistently projected to grow 15-28%.

Anthropic Economic Index (January 2026)

→ 2026

Anthropic's January 2026 Economic Index report measures actual Claude API usage by occupational task category. Legal occupations show a notable gap between theoretical AI capability (~80% of tasks theoretically addressable) and observed actual usage (~15% of tasks actively delegated to Claude in production). Paralegals, legal researchers, compliance analysts, and business intelligence analysts are specifically cited as having high exposure — with tasks like document review, contract summarization, research synthesis, and report drafting being handled by Claude. Reported here as -15% on the cone to represent *current observed adoption* rather than theoretical ceiling; the gap between 80% capability and 15% observed adoption reflects regulatory caution, professional liability concerns, and ABA ethics rules rather than technical limitations.

McKinsey Global Institute (July 2023)

→ 2030

McKinsey's July 2023 "Generative AI and the Future of Work in America" report projects that generative AI could automate up to 30% of current US work hours by 2030 — with legal and business professional services among the sectors with both high automation potential *and* high demand growth. McKinsey's net view for legal occupations is more nuanced than Frey/Osborne or Goldman: they model legal workers as benefiting from productivity gains even as some tasks automate, suggesting paralegals who upskill into AI-augmented workflows may see demand hold flat rather than collapse. The -30% figure represents task-automation potential applied to this role, not a projected net employment change.

Goldman Sachs — "Generative AI and Legal Work" (March 2023)

→ 2030

Goldman Sachs March 2023 report "Potentially Large Effects of AI on Economic Growth" identified legal occupations as having 44% of tasks automatable by current generative AI — the highest sector-level exposure in the entire study, above even administrative work (46% is the admin figure, legal is cited in the report as approximately 44% at the sector level by economist Jan Hatzius). Note: this is share of *tasks* automatable, not share of jobs lost — the report projects 300M global jobs affected across all sectors but does not disaggregate to individual occupation headcounts.

Eloundou et al. — "GPTs are GPTs" (2023)

→ 2024

GPT-4 task-by-task labeling against O*NET task statements for 23-2011. Legal occupations overall score high on Eloundou's β (E1 + 0.5×E2) metric — conventional secondary-source citation places paralegals and legal assistants among the highest-exposure professional occupations in the dataset, with γ (any exposure) near 0.76 and β near 0.50-0.60. Reported here as -76% to represent γ on the cone display; β would be lower (~-55%). "Exposure" is *capability*, not substitution — it measures the share of tasks an LLM could assist with, treating this as a ceiling on displacement rather than a floor.

Frey & Osborne (2013)

→ 2033

Gaussian-process classifier on O*NET task features. Frey & Osborne classified paralegals and legal assistants at 0.94 probability of computerisation — among the highest of all 702 occupations studied, often cited as ranking approximately 6th most-automatable in the entire set. The high score reflects the largely rules-based, information-retrieval nature of the tasks: Shepardizing, case research, document indexing, and form preparation all scored high on the bottleneck criteria of "fine motor skills" (low), "creative intelligence" (low), and "social intelligence" (low). Note: this is not a 94% employment-loss forecast — it is the probability that the *core task set* is fully automatable within ~20 years; the actual employment effect depends on adoption speed, regulatory friction, and whether demand expands.