YourJobs AI Vulnerability Index — Methodology
By YourJobs Editorial · Version 1.0 · Published · Last updated
The YourJobs AI Vulnerability Index is a 0-100 composite score quantifying how exposed each US occupation (SOC 2018 detailed code) is to displacement or significant task reshape by current-generation AI within five years. Lower scores indicate roles where human judgement, physical interaction, or interpersonal trust dominate. The score combines three published research signals — Frey & Osborne (2013), OpenAI/Penn (2023), and the Anthropic Economic Index (2024) — into a single ranking, refreshed annually as new evidence emerges.
Formula
The composite score is a weighted average of three sources, scaled to a common 0-100 range and rounded to integer. Forty per cent weight goes to peer-reviewed academic exposure (Frey & Osborne 2013), forty per cent to LLM-era task analysis (OpenAI/Penn 2023), and twenty per cent to observed-in-practice usage (Anthropic Economic Index 2024). The weighting balances theoretical exposure against measured adoption.
score = round(0.40 * frey_osborne_2013 + 0.40 * openai_penn_2023 + 0.20 * anthropic_economic_index_2024)
Components
Each input is a published per-occupation score, scaled to 0-100 and weighted as follows. Click any source to read the original study.
Frey & Osborne — automation probability (weight 0.4)
Source: https://www.oxfordmartin.ox.ac.uk/downloads/academic/future-of-employment.pdf
Scale: 0-100 (probability of computerisation × 100)
Estimates the probability of full computerisation by ~2030. Predates LLMs but remains the most cited mapping. Scaled directly.
OpenAI / UPenn — GPTs are GPTs (weight 0.4)
Source: https://arxiv.org/abs/2303.10130
Scale: 0-100 (β fully-exposed-share × 100)
Per-occupation share of tasks where access to LLMs reduces task time by ≥50% without quality loss. Captures LLM-era exposure that Frey & Osborne missed.
Anthropic Economic Index (weight 0.2)
Source: https://www.anthropic.com/news/the-anthropic-economic-index
Scale: 0-100 (relative claude-usage share × 100, normalised)
Empirical measurement of which occupations actually receive Claude usage. Captures observed-in-practice exposure.
Score Bands
The 0-100 range is partitioned into four interpretive bands. The bands are not equal-width; they reflect plateaus where qualitative interpretation changes. A score near a band boundary should be read as "either-or" rather than "definitely the upper band".
- Low (0–25): Tasks are largely physical, interpersonal, or judgement-heavy in ways current AI cannot replicate.
- Moderate (26–50): Some tasks are AI-augmentable; near-term role evolution rather than replacement.
- High (51–75): Significant share of routine work is AI-augmentable; structural pressure on headcount within 5 years.
- Very High (76–100): Most tasks are within current AI capability; aggressive role redesign expected within 3 years.
Limitations
Each input source has documented limitations, and the composite inherits them. The index is approximate guidance, not a forecast for any individual role.
- Frey & Osborne predates LLMs (2013) — overweights manual automation; underweights cognitive automation.
- OpenAI/Penn uses task-level analysis; aggregating to occupation introduces estimation error.
- Anthropic Economic Index reflects Claude usage share, not total LLM exposure across providers.
- All three sources are US-centric; international applicability untested.
- Composite scores are approximate guidance, not predictions for any individual role.
Frequently asked
What is the YourJobs AI Vulnerability Index?
A 0-100 composite score per US occupation (SOC 2018 detailed code) quantifying how exposed the role is to displacement or significant task reshape by current-generation AI within five years. Lower scores indicate roles where physical interaction, interpersonal trust, or accountable judgement dominate. Higher scores indicate roles where a substantial share of tasks is already within AI capability.
What does an AI score of 70 mean?
A score of 70 places the occupation in the "High" exposure band (51-75). Roughly speaking, a substantial share of routine work in the role is AI-augmentable, and structural pressure on headcount is plausible within five years. It is not a forecast that the role will disappear; it is a measurement of how much of the work could be done by current AI given access.
Are these scores predictions of which jobs will be lost?
No. The index measures *current AI capability* against the known task structure of each occupation, not labour-market outcomes. Whether a role actually shrinks depends on regulation, capital allocation, demand growth, productivity gains being passed to consumers, and many other factors the index does not model. Treat the score as a capability map, not a forecast.
How often is the index updated?
Annually, when new evidence from the source studies is published. Component weights and source data are reviewed each January. The next planned refresh incorporates the OpenAI/Penn 2026 update and the latest Anthropic Economic Index release.
Why these three sources?
Each source captures a different facet of AI exposure that the others miss. Frey & Osborne (2013) provides peer-reviewed automation probability across 702 occupations — the most cited mapping in the field, but it predates large language models. OpenAI/Penn (2023) measures LLM-era task exposure that Frey & Osborne missed. The Anthropic Economic Index (2024) measures observed-in-practice Claude usage, capturing real workplace adoption rather than theoretical exposure.
Can I cite or reproduce these scores?
Yes. The dataset is published under Creative Commons Attribution 4.0. Cite as: "YourJobs AI Vulnerability Index (2026). Methodology v1.0. https://yourjobs.com/ai-index-methodology". Direct links to individual occupation pages are also welcome.
Citation
If you reference these scores, please cite:
YourJobs AI Vulnerability Index (2026). Methodology v1.0. https://yourjobs.com/ai-index-methodology
The dataset is published under Creative Commons Attribution 4.0. Attribution required; commercial reuse permitted.