What Jobs Are Safe From AI? [March 2026]

🟢 The Safety Spectrum

AI safety is not a binary switch. It’s a spectrum. Every role sits somewhere between “fully protected” and “fully exposed,” determined by structural characteristics that either block or invite automation. Understanding where your role sits — and why — is the first step toward making an informed career decision.

The spectrum is not random. It maps directly to structural characteristics. Roles on the safe end require a human body in a physical location, a legal licence to practise, or a trust relationship that people won’t accept from a machine. Roles on the exposed end are primarily screen-based, pattern-matching tasks with no licensing requirement.

The three zones represent fundamentally different relationships with AI technology. GREEN zone workers use AI as a tool that makes them more productive — a nurse using AI-assisted diagnostics sees more patients, a cybersecurity analyst using AI threat detection catches more attacks. YELLOW zone workers share tasks with AI — some parts of their job are automated, but human oversight remains essential. RED zone workers find that AI can perform their core deliverables — writing, data processing, translation, basic coding — independently.

Understanding the spectrum matters because it changes how you should respond. GREEN zone: embrace AI tools to boost your productivity. YELLOW zone: upskill to own the parts AI can’t do. RED zone: consider transitioning toward roles with structural protection. All three are valid — but only if you know which zone you’re in.

The Three Types of Safety

Not all safe jobs are safe in the same way. The GREEN zone contains three sub-categories, each with a different relationship to AI:

The sub-category matters for how you should respond. Accelerated: lean in — AI is your tailwind. Stable: keep doing what you’re doing — the barriers are permanent. Transforming: learn the AI tools for your field — the role is safe but the methods are changing.

🏆 Top 20 Safest Jobs From AI

These are the 20 highest-scoring roles in our database. Each one combines multiple structural barriers that AI cannot overcome: physical presence, licensing, trust, and real-time judgement. These aren’t predictions — they’re measurements of structural protection that exists today.

The pattern in the top 20 is unmistakable: physical presence (surgeons, electricians, firefighters), regulatory licensing (doctors, nurses, pilots), and trust relationships (therapists, teachers, social workers). Most top-20 roles have all three barriers. The average score across the top 20 is well above 70 — deep in GREEN zone territory, with multiple reinforcing layers of protection.

Showing the top 20 of 1769 GREEN zone roles. View all 1769 →

The contrast is instructive. Every trait that makes a role AI-safe — physical presence, licensing, trust, variable conditions — is also a trait that makes it harder to fill. That’s why the safest jobs face persistent shortages: the same barriers that block AI also limit the supply of qualified humans. For workers, this is the best possible position: protected from automation AND in high demand.

The Safety Boundary: Where GREEN Meets YELLOW

Safety isn’t binary. Roles scoring 48+ are GREEN — structurally safe. But roles just below that threshold, in the YELLOW Moderate zone, are still relatively protected. They face more AI augmentation than GREEN roles, but replacement isn’t the concern. The difference between a score of 46 and 50 is real, but it’s a spectrum, not a cliff.

Top YELLOW Moderate Roles

The highest-scoring YELLOW Moderate roles — still safe, but with more AI exposure than GREEN zone jobs.

🛡️ What Makes a Job Safe From AI?

Four structural barriers determine whether a job is safe from AI. No amount of AI capability improvement can overcome a requirement for a human body, a legal licence, an emotional connection, or real-time decisions in chaotic environments. Roles with multiple barriers are the safest. Roles with none are the most exposed.

Why can’t AI overcome these barriers? Physical presence is a hardware problem: even the most advanced robots can’t match human dexterity in variable environments (a pipe in a different position every time, a patient in a different condition every time). Licensing is a legal problem: legislatures move in years, not months, and no electorate will vote to let AI practise medicine unsupervised. Trust is a psychological problem: humans form relationships with other humans — a therapist’s effectiveness depends on the patient believing they’re heard by a person.

These barriers aren’t temporary limitations that will be solved with better models. They’re structural — rooted in physics (bodies in space), law (regulatory frameworks), and human psychology (trust and empathy). An LLM that’s 10x more capable still can’t wire a house. An AI agent that’s 100x more capable still can’t legally prescribe medication. The barriers are not about AI intelligence — they’re about what a digital system fundamentally cannot do.

🤖 What AI Can vs Cannot Do

Assessing your own role’s safety starts with understanding what AI actually does well and where it fundamentally fails. The gap is not closing uniformly — AI advances fast in digital pattern-matching but makes zero progress on physical presence, legal authority, or genuine human connection.

The key insight: AI’s capabilities are improving rapidly within the “can do” column. But the “cannot do” column is not about AI intelligence — it’s about what a digital system fundamentally is. An LLM that’s 1,000x more capable still has no hands, no licence, no legal standing, and no genuine emotional experience. These are not limitations that will be “solved” with better models.

The “cannot do” list is not shrinking. AI researchers have made extraordinary progress in text, code, image, and video generation. But progress on physical manipulation in variable environments is minimal. Progress on legal standing is zero — no legislature is seriously considering granting AI practitioner licences. Progress on genuine empathy is conceptually impossible for a system without lived experience. The safe side of the spectrum is structurally stable.

This distinction also explains why “AI will take all jobs” is wrong and “AI won’t affect any jobs” is equally wrong. AI is extremely capable in specific domains and completely incapable in others. The question for your career is which side your core deliverables sit on. If you fix pipes, examine patients, teach children, or protect networks, you’re on the safe side. If you process data, write reports, or categorise information, you’re on the exposed side.

🏭 Safety by Industry

AI safety clusters by sector. Healthcare, trades, education, and cybersecurity dominate the GREEN zone. Finance and administration sit at the opposite end. The domain scores reveal which industries offer structural protection and which leave workers exposed.

The domain scores reveal a structural divide: industries built on physical work, licensing, and human relationships score highest. Industries built on digital, pattern-based work score lowest. This isn’t opinion — it’s the mathematical result of scoring 3649 roles across five AI resistance dimensions.

The pattern is clear: if you’re choosing between industries for AI safety, pick one that scores above 50. Healthcare, trades, education, cybersecurity, and engineering all offer structural protection. The specific role within those industries matters too — but starting in a high-scoring domain gives you a strong foundation.

For career planners, the domain scores serve as a quick filter. Before researching specific roles, check whether the entire sector scores well. A role in a high-scoring domain has tailwinds — even if it’s not the highest scorer, it benefits from the structural characteristics that protect the whole sector. A role in a low-scoring domain faces headwinds — even strong individual performers may be affected as the sector transforms.

🏥 Healthcare & Nursing

Healthcare is the single most AI-resistant major sector. The WHO projects a 10 million health worker shortage by 2030 — demand is growing faster than AI can augment, let alone replace. Nurse practitioners, registered nurses, surgeons, and therapists all sit firmly in the GREEN zone. Physical examination, licensing, and patient trust create triple-layer protection.

Healthcare roles are protected by triple barriers: physical examination, licensing, and patient trust. AI tools assist with diagnostics and record-keeping, but no jurisdiction permits AI to independently examine a patient, prescribe treatment, or perform surgery. The sector faces shortages, not surplus. Nurse practitioners alone are projected to grow 45% — faster than nearly any other occupation.

The healthcare data is unambiguous across every source we track. The US needs 86,000 more physicians by 2034 (AAMC). The NHS has 107,000 unfilled positions. These aren’t projections about what might happen — they’re measurements of gaps that exist right now and are widening. If you’re in healthcare, your role is not just safe from AI — it’s one of the most structurally secure positions in the entire economy.

AI is transforming healthcare workflows — diagnostic imaging, drug discovery, patient scheduling, medical records. But every one of these AI applications makes clinicians more effective, not redundant. A radiologist using AI reads scans faster and more accurately. A nurse using AI patient management spends less time on paperwork and more time at the bedside. AI in healthcare is the clearest example of augmentation, not replacement.

The mental health sector deserves special attention. BLS projects 19% growth for mental health counsellors. The WHO reports a global mental health workforce crisis. Demand is driven by rising awareness, reduced stigma, and pandemic-era trauma. AI chatbots exist for mental health support, but research consistently shows the therapeutic alliance — the relationship between therapist and client — is the strongest predictor of treatment outcomes. A human therapist isn’t just preferred — they’re clinically more effective.

🔧 Trades & Construction

Skilled trades are the most structurally protected occupation group in the economy. AI cannot wire a house, fix a pipe, or pour concrete — and there is no timeline where it can. The AGC reports 91% of US construction firms struggle to fill positions. These roles require physical presence, licensing, and hands-on dexterity — the exact combination that makes them immune to AI displacement.

The trades are the clearest example of structural AI safety. Every role requires physical presence on a job site. Most require professional licensing or certification. The work is variable — no two wiring jobs, plumbing repairs, or construction sites are identical. Infrastructure spending ($1.2T from the IIJA alone) is adding demand on top of existing shortfalls. These roles don’t just survive AI — they thrive alongside it.

The trades also have a demographic crisis: the average construction worker is ageing out. The AGC reports that recruitment of younger workers isn’t keeping pace with retirements. This creates a double demand signal: replacement of retiring workers PLUS new demand from infrastructure spending. For anyone entering the trades now, the supply-demand dynamics are the most favourable they’ve been in a generation.

The trades also demonstrate why AI safety is not about intelligence or education level. An electrician’s work requires deep technical knowledge (national electrical codes, circuit design, load calculations), practical skill (pulling wire through walls, terminating connections in tight spaces), and professional judgement (diagnosing faults, ensuring safety in live environments). The salary data proves it: a master electrician with 10 years of experience can earn over $80,000, with zero student debt.

Construction technology is advancing rapidly — 3D printing, drone surveys, BIM modelling, IoT sensors — but every advancement creates demand for the humans who operate, maintain, and oversee these systems on job sites. A 3D-printed wall still needs an electrician to wire it. A drone survey still needs an engineer to interpret it. The technology makes the work more sophisticated, not less human-dependent.

🎓 Education & Teaching

Teaching requires physical classroom presence, trust relationships with students and parents, and real-time judgement in unpredictable environments. UNESCO estimates the world needs 44 million additional teachers by 2030. AI tools help teachers work more efficiently, but they don’t replace the human in the room.

Teaching requires physical classroom presence, real-time adaptation to student needs, and trust relationships with children and families. UNESCO estimates 44 million additional teachers needed by 2030. The US alone has teacher shortages in 47 states. AI tools assist with lesson planning and grading but cannot replace the teacher in the room.

The education sector highlights a critical point about AI augmentation vs replacement. AI tools are transforming how teachers work — automated grading, personalised learning platforms, AI-generated lesson plan suggestions. But none of these replace the teacher. They make the teacher more effective. A teacher using AI grading tools can spend more time on one-to-one student interaction — the part of the job that matters most and that AI cannot do.

Special education is the strongest example within teaching. SPED teachers work with students who have unique, unpredictable needs requiring moment-to-moment professional judgement, physical assistance, and deep trust relationships with students and families. 45 US states report SPED as a shortage area. AI cannot observe a child’s non-verbal cues, adjust a lesson in real-time, or build the trust a non-verbal student needs to engage.

🔒 Cybersecurity

Cybersecurity is the paradox sector: AI creates more security jobs, not fewer. Every AI system deployed creates new attack surface. ISC2 reports a global workforce gap of 4.8 million. BLS projects 33% growth for information security analysts through 2033 — six times the national average. This sector grows in direct proportion to AI adoption elsewhere.

Cybersecurity is unique: AI increases demand for human security professionals. Every AI system deployed creates new attack surface. Every automated process introduces new vulnerability. The ISC2 reports a 4.8M workforce gap that is widening, not closing. Cybersecurity analysts earn a 16% salary premium over general IT — and that gap is growing as demand outstrips supply.

Cybersecurity also benefits from the regulatory tailwind. GDPR, the EU AI Act, NIS2, and expanding US compliance requirements all mandate security practices that require human oversight. As regulation increases, so does demand for compliance-trained security professionals. The combination of expanding attack surface, regulatory requirements, and criminal innovation makes cybersecurity the one sector where AI adoption directly increases human employment.

Fortinet reports that 87% of organisations that experienced a breach in the past year attributed it at least partly to a cyber skills gap. The talent shortage is not abstract — it has direct, measurable consequences for organisations. This creates an urgency in hiring that benefits every worker in the sector. Employers are competing for cybersecurity talent, not the other way round.

The cybersecurity career path also offers unusually strong upward mobility. Entry-level SOC analyst positions lead to specialisation in penetration testing, cloud security, incident response, or security architecture — each commanding progressively higher salaries. The ISC2 reports that cybersecurity professionals earn a 16% premium over general IT roles at every career stage. Combined with AI-driven demand growth, this makes cybersecurity one of the most financially rewarding AI-safe career paths available, especially for career changers from other technical fields.

⚙️ Engineering

Engineering combines advanced education, professional licensing, and physical-world application. Civil engineers visit sites. Electrical engineers work on live systems. Environmental engineers assess real terrain. The IET reports the UK alone needs 124,000 additional engineers and technicians annually. AI augments design but cannot replace the licensed professional in the field.

Engineering combines advanced education, professional licensing, and physical-world application. AI augments the design process (CAD, simulation, optimisation) but cannot replace the licensed engineer who signs off on the bridge design, inspects the wiring, or assesses the environmental impact on-site. The median engineer wage is well above the national average, reflecting both skill scarcity and structural demand.

The semiconductor reshoring trend is adding another demand layer. As countries build domestic chip fabrication capacity (CHIPS Act in the US, similar programmes in the EU and Asia), demand for electrical, chemical, and industrial engineers is surging. Infrastructure electrification (EV charging, grid modernisation, heat pump adoption) adds further demand for electrical engineers specifically. The IET reports 124,000 engineers and technicians needed annually in the UK alone — a gap that has been widening for a decade.

Civil engineering is particularly interesting from an AI-safety perspective. Every civil engineering project involves unique site conditions, regulatory requirements, community impact assessments, and physical inspections. AI can optimise a bridge design in simulation, but the civil engineer must visit the site, assess the terrain, review the geology, and sign off on the plans with their professional licence. That chain of physical presence + licensing + professional judgement is unbreakable by AI.

🌿 Green Economy & Energy

The energy transition is creating AI-resistant jobs at industrial scale. Wind turbine technicians and solar installers are the two fastest-growing occupations in the US economy. Both require physical presence, specialised training, and work in unpredictable outdoor environments. IRENA reports 16.2 million renewable energy jobs worldwide.

Wind turbine technician (+60%) and solar installer (+48%) are the two fastest-growing occupations in the US economy. Both require physical presence at outdoor sites, specialised training, and work in variable conditions that AI cannot navigate. The IEA projects 35 million clean energy jobs globally by 2030. The energy transition is building an entirely new category of AI-safe employment.

The green economy also creates demand for existing AI-safe trades: electricians install EV chargers, HVAC technicians fit heat pumps, construction crews build solar farms and wind sites. The energy transition doesn’t just create new green jobs — it amplifies demand for traditional trades that are already in shortage. The US DOE reports 3.4 million clean energy jobs domestically, with the sector growing faster than the overall economy.

For career planning, the green economy represents the intersection of three powerful trends: AI resistance (physical, outdoor work), policy support (government spending and incentives), and structural demand (the energy transition is a multi-decade project). Workers entering this sector now have a 30+ year runway of guaranteed demand ahead of them.

💰 Safe Jobs Pay Well

The roles AI cannot replace aren’t just safe — they pay well. BLS data shows GREEN zone sectors consistently offer median salaries 20–40% above the national average. The same barriers that block AI (licensing, physical skill, advanced training) also limit supply, which drives wages up.

The salary data tells the shortage story in numbers. Nurse practitioners earn $126K. Cybersecurity analysts earn $120K. Construction managers earn $105K. Engineers average $98K. These are all significantly above the US median of $48K. When demand exceeds supply, wages rise — and every GREEN zone sector shows above-average wage growth. Being AI-safe and being well-paid are not just correlated — they’re causally linked.

The salary data demolishes the myth that “safe” jobs are low-paid jobs. The highest-scoring roles in our database are also among the highest-paid. This is not a coincidence — it’s economics. The same barriers that block AI (licensing, physical skill, advanced education) also limit the supply of qualified workers, which drives wages up. Being AI-safe and being well-compensated are two expressions of the same underlying scarcity.

For workers currently in lower-paying exposed roles, this data reframes the career transition calculation. Moving from a RED zone job paying $40K to a GREEN zone trade paying $62K (electrician) or a GREEN zone healthcare role paying $75K+ (registered nurse) is not just an AI-safety move — it’s a significant income upgrade. The retraining investment typically pays for itself within 2–3 years through higher earnings alone, independent of the AI protection benefit.

⚠️ The Shortage Advantage

The global talent shortage in AI-safe sectors is not an anecdote — it’s a measured, persistent, and worsening trend. ManpowerGroup reports 74% of employers worldwide struggle to find skilled workers. The shortage is concentrated in exactly the sectors that score highest: healthcare, trades, education, and engineering.

Korn Ferry projects an 85 million worker talent deficit by 2030, potentially costing $8.5 trillion in unrealised revenue. The shortage is concentrated in exactly the sectors that score highest: healthcare (10M gap), cybersecurity (4.8M gap), teaching (44M needed), engineering (124K/year in the UK alone), and trades (91% of construction firms can’t fill roles). These aren’t just safe careers — they’re careers where the world is begging for more workers.

The shortage advantage is perhaps the most underappreciated aspect of AI-safe careers. Everyone talks about which jobs AI will replace. Almost nobody talks about the fact that the safest jobs are simultaneously the ones with the biggest talent gaps. This creates a career sweet spot: structurally protected from AI, actively sought by employers, and offering rising wages. It’s the best labour market position available in 2024 and the data says it will only improve through 2030.

The shortage data also has implications for job seekers’ negotiating power. In sectors where 74% of employers struggle to find talent (ManpowerGroup), workers have leverage. This translates into higher starting salaries, better benefits, more flexible schedules, and employer-funded training and certification. The power dynamic in AI-safe sectors favours the worker — a stark contrast to AI-exposed sectors where automation gives employers leverage to reduce headcount and compress wages. Choosing an AI-safe career is not just about avoiding displacement — it’s about choosing the side of the labour market where workers have structural bargaining power.

🔍 How to Check If Your Role Is Safe

Knowing the theory is not enough. You need a practical framework to assess your own role. Below is a structured self-assessment you can apply to any job — current or prospective — to determine where it sits on the AI safety spectrum.

The self-assessment works because it maps directly to the structural barriers that determine AI resistance. Every GREEN zone role in our database scores highly on questions 1–5. Every RED zone role scores low. The questions aren’t arbitrary — they measure the same dimensions our scoring framework uses to classify 3649 roles.

What to Do Based on Your Result

Reskilling Pathways Into Safe Careers

If you’re considering a transition to a safer role, most GREEN zone sectors have faster entry routes than people assume. Many don’t require a four-year degree.

✅ The Bottom Line

Is your job safe from AI? The data gives a clear answer. 1769 of 3649 roles score GREEN on the JobZone Index, covering 56.2M US workers. They cluster in healthcare, trades, education, cybersecurity, and engineering. They pay above-average wages. They face persistent shortages in every developed economy. And they share structural barriers — physical presence, licensing, trust, and unpredictable environments — that AI cannot overcome on any foreseeable timeline.

This isn’t optimistic guesswork. It’s the mathematical result of scoring 3649 roles against real AI capabilities and cross-referencing with 63+ externally-sourced statistics from the WHO, BLS, WEF, ManpowerGroup, ISC2, UNESCO, IRENA, and more. The data from every source tells the same story: the jobs that require human bodies, human licences, and human trust are safe. The jobs that are primarily digital pattern-matching are not.

This page is updated as new role assessments are added and external data refreshes. The structural patterns don’t change — physical work, licensing, and trust are permanent features of the economy, not temporary ones — but the specific numbers, shortage figures, and salary data are updated regularly.

For the opposite perspective — which jobs AI will replace — see What Jobs Will AI Replace First? For a comprehensive look at job loss data, see AI and Job Loss Statistics. For the full list of roles AI cannot replace, see Jobs That AI Cannot Replace.

The question “is my job safe from AI?” has a clear, data-backed answer. Check your structural barriers. Count the physical, licensing, and trust elements. The more you have, the safer you are. And if you’re not safe, the data also shows you exactly where to go: the sectors with the biggest shortages, the highest wages, and the strongest structural protection. Search all 3649 assessed roles to find where yours sits.