Are Warehouse Workers Safe from AI? The 2026 Risk Picture
Warehouse workers and logistics labourers (ISCO 93) score 2.0/10 on AI exposure - very low. Sorting, picking, packing, and stacking are physical tasks, not language tasks. The real threat is robotics, where the score rises to 6.5/10. Amazon Robotics reported 750,000 robots deployed across its fulfilment network in 2025. ILO ILOSTAT estimates 50 million warehouse and storage workers globally.
Key findings
- Warehouse workers (ISCO 93): AI exposure 2.0/10 - very low; physical tasks don't translate to language model risk
- Robotics risk 6.5/10 - high and accelerating (Amazon: 750k robots, 2025)
- 1.2M US hand labourers and material movers, median $36,980/yr (BLS OEWS May 2025)
- 50M global warehouse and storage workers (ILO ILOSTAT 2024 estimate)
- Safest roles: non-standard item handling, returns processing, team supervision
Why warehouse workers score low on AI but high on robotics
Large language models and AI systems are defined by their ability to process and generate text. Sorting boxes, picking items from shelves, operating pallet jacks, wrapping pallets, and loading trucks are physical coordination tasks in three-dimensional space. They require no language output - they require hands, spatial awareness, and the ability to handle irregular physical objects under time pressure. AI exposure scores reflect the proportion of tasks within an occupation that are language, prediction, or classification tasks. For ISCO 93 warehouse labourers, that proportion is very low - hence 2.0/10.
Robotics risk measures something different: whether the physical motions of a job can be performed by a robotic system with sufficient precision and at sufficient cost-effectiveness to justify replacement. For warehouse work, this measure is significantly higher. The ILO ILOSTAT-derived ISCO 93 robotics score of 6.5/10 reflects a structured, predictable physical environment with repetitive motions - precisely the conditions that robotic systems handle most effectively.
This matters for workers and employers alike. A warehouse worker preparing for the future should not be studying AI. They should understand that the displacement risk - if it materialises - comes from industrial robots, autonomous mobile robots (AMRs), and automated conveyor and sorting systems, not from ChatGPT.
Amazon Robotics: the scale of deployed automation
| System | Function | Scale (2025) | Robotics risk impact |
|---|---|---|---|
| Amazon Robotics (Kiva-derived AMRs) | Shelf transport to picker | 750,000 units deployed | High |
| Sparrow robotic arm | Item picking from totes | Selected FC sites | High |
| Cardinal robotic arm | Package sorting by destination | Scaling deployment | High |
| Proteus AMR | Cart transport (human-collaborative) | Pilot FCs | Medium |
| Automated conveyor and scan systems | Inbound receiving and outbound sort | Standard across large FCs | Medium |
Source: Amazon Annual Report 2024; Amazon Robotics press releases 2024-2025. Amazon's own stated goal is the "hybrid" model - humans and robots working in the same fulfilment centre - not full automation. This reflects both the technical limitations of current robotic picking (success rates for unstructured items remain below human performance) and the operational reality that irregular items, returns, and exceptions still require human judgment.
Warehouse role breakdown by automation risk
| Role | AI score | Robotics score | Automation status |
|---|---|---|---|
| Standard item picking (flat-path AMR) | 2.0/10 | 8.0/10 | Partially automated, scaling |
| Pallet stacking and wrap | 2.0/10 | 6.0/10 | Partially automated |
| Outbound sort and load | 2.0/10 | 6.5/10 | Partially automated |
| Inbound receiving and checking | 2.5/10 | 5.0/10 | Partially automated |
| Returns processing with damage assessment | 3.0/10 | 4.0/10 | Human-led; AI supports disposition |
| Non-standard and fragile item handling | 2.0/10 | 4.0/10 | Human-led; irregular geometry challenges robots |
| Warehouse team leads and floor supervisors | 2.5/10 | 3.0/10 | Human-led; oversight, exception handling |
Why full warehouse automation is still 5-10 years away in large markets
The major technical blocker is robotic grasping of unstructured items. An Amazon fulfilment centre handles items of every possible shape, weight, and fragility. Standard AMRs successfully transport shelves to human pickers. Robotic arms successfully pick predictable items from structured totes - but their success rate on the full SKU range that an Amazon FC stocks remains below human performance as of 2025. Sparrow, Amazon's item-picking arm, was operational at a subset of sites as of the 2024 Annual Report, not deployed fleet-wide.
The economic calculus also limits speed. At $36,980 median annual wage (BLS OEWS May 2025), a US warehouse worker costs less per year than many robotic systems cost to deploy and maintain at that station. The cost-crossover point - where the robot is cheaper over a 3-year depreciation cycle - is arriving for standard-item picking but has not arrived universally.
In India, Vietnam, Bangladesh, and most of sub-Saharan Africa, the labour cost differential makes warehouse robotics economically marginal for a decade or more. ILO ILOSTAT 2024 data shows that the largest concentrations of ISCO 93 workers are in South and East Asia and Sub-Saharan Africa - populations where the 6.5/10 robotics risk score is a horizon measure, not a near-term threat.
See warehouse and logistics risk by country
ISCO 93 worker concentration and robotics deployment pace vary significantly across 206 countries. Compare warehouse automation risk in your country.
Open the Explore Tool →Was this analysis useful?