Prison Guards

← The GenAI exposure gradient

On the International Labour Organization's 2025 global study, the 7 task statements that define Prison Guards (ISCO-08 5413) score an average of 0.14 on a 0–1 exposure scale — more exposed than about 13% of the 427 placed occupations. Roughly 0% of its tasks fall somewhere on the exposed part of the gradient, and the typical task lands in the Not exposed band.

Exposure is task overlap, not a verdict. A high score means a generative-AI model can do part of the content of these tasks — it says nothing about whether the work is automated, whether anyone uses AI for it today, or whether jobs are lost. The gradient is scored on the international ISCO-08 system; the rest of Singulariki is U.S. O*NET/SOC, bridged below by an approximate, many-to-many crosswalk.

How its tasks split across the gradient

Each of the 7 scored tasks for this occupation, sorted into the six exposure bands — cool (human ground) to hot (almost fully assistable).

The most-exposed task

“Assisting with the implementation of rehabilitation programmes;”

Scores 0.21 on the 2025 scale. The task of assisting with the implementation of rehabilitation programs involves a blend of human interaction, empathy, and understanding, which are challenging for Generative AI to replicate fully. This is similar to tasks such as "Directing early rehabilitation of patients after surgical procedures" (adjusted score of 0.18) and "Planning the rehabilitation of hearing and speech for deaf or hard-of-hearing individuals" (adjusted score of 0.2), both requiring significant human judgment and adaptation to patient-specific conditions. While AI can assist by providing data-driven insights, generating activity suggestions, and organizing schedules, the core element of personalization and interaction with patients remains a human-centered task. The role of AI in this context is supportive, aiding with documentation or providing standardized information, but not autonomously implementing rehabilitation plans. Considering these factors and the technological infrastructure in high-income countries like Poland, an adjusted score of 0.18 reflects the limited but existent potential for AI assistance in this task, emphasizing the essential human component.

Moving fastest, 2023 → 2025

“Supervising prisoners at work and meals, and during recreation periods;”

Model capability on this task changed by +0.07 in two years — the gradient is not static, it is filling in.

U.S. occupations this maps to

The American O*NET/SOC roles that crosswalk to ISCO-08 5413, biggest by employment first, via the published (approximate, many-to-many) IBS O*NET-SOC ↔ ISCO-08 correspondence. These are the closest U.S. matches — not an asserted one-to-one identity.

• Correctional Officers and Jailers
• First-Line Supervisors of Correctional Officers

In context

Part of the 5 - Service and sales workers major group. Return to the full gradient to see how the whole group sits.

Prison Guards sit at the 13th percentile of the global GenAI exposure gradient

• Across 427 international occupations scored by the ILO, Prison Guards rank in the 13th percentile for GenAI task exposure — overlap with what generative AI can attempt, not a projection of displacement. (ILO / Gmyrek et al. (2025) GenAI exposure gradient)
• About 0% of this occupation's tasks fall into an exposed gradient band. (ILO / Gmyrek et al. (2025))
• Mean task exposure rose by 0.01 between the 2023 and 2025 model-capability snapshots. (ILO / Gmyrek et al. (2025), 2023→2025)
• Its most-exposed task: "Assisting with the implementation of rehabilitation programmes;". (ILO / Gmyrek et al. (2025))

Source: Singulariki — "Prison Guards". https://singulariki.com/gradient/5413-prison-guards.html
Note: AI task overlap measures what today's AI can attempt, not automation, job loss, or a forecast.