The prospect of a job apocalypse, blighting the career prospects of today’s wave of college graduates, is concerning policy-makers and parents alike. But surely there are some fields that will prove resilient? I wonder what Claude thinks…

It might seem ironic to be quizzing an LLM about career domains that will prove resistant to the changes being wrought by… LLMs.

Nevertheless, when Pete Andreasen asked Claude, the Anthropic LLM to predict those career fields that offered the best prospects of continued employment, here’s what Claude had to say.

The core principle that determines protection

Before naming fields, it’s worth understanding why some roles resist automation. The pattern is consistent across all research: if a job is predictable and rule-based, AI can automate it. It’s not the job title — it’s the skill stack.

The most protected roles share a cluster of traits: they require high-stakes judgment where errors have real-world consequences, involve adversarial or unpredictable environments that evolve faster than models can be trained, require accountability that can’t be delegated to a machine, or sit at the nexus of human trust and technical complexity.

With that framework, here’s where the evidence actually points.

1. Cybersecurity — the strongest structural case

This is the single most research-consistent answer across sources, and it has a logic that goes deeper than other fields. The Bureau of Labor Statistics estimates 33% growth in cybersecurity roles such as Information Security Analysts by 2033, making it one of the most future-proof IT careers.

But the more important reason is structural: AI makes the attack surface larger and more sophisticated, not smaller. As AI increases digital capabilities, it also increases vulnerabilities. These roles exist precisely because automation creates new risks. It’s an adversarial arms race — every AI tool deployed in defense is matched by AI tools used by attackers, and that dynamic guarantees human demand.

AI helps detect anomalies — but humans decide response strategies.

Within cybersecurity, the highest-protection specializations are shifting. The rise of autonomous agents has made traditional “scan and patch” models partially obsolete, with the industry pivoting toward an “Autonomy vs. Autonomy” paradigm where AI security and AI red-teaming are the growth edge. AI red-teamers — people who probe AI systems for exploitable weaknesses, prompt injection vulnerabilities, and adversarial attacks — are a genuinely new role that didn’t exist five years ago and that AI itself cannot fill.

AI security is currently the highest-paying specialization in cybersecurity, with a massive talent gap because traditional security pros have not upskilled yet.

2. AI/ML Engineering and Infrastructure — the field that is AI

AI engineering, infrastructure architecture, and AI safety and alignment are seeing the lowest layoff rates and highest demand across all the tracked 2026 layoff data. This seems obvious but deserves unpacking.

The people building, training, fine-tuning, and deploying AI systems are not being replaced by AI — they’re being hired at a premium to build the thing replacing everyone else. The catch is that this field requires a genuinely high bar: strong mathematics (linear algebra, probability, optimization), systems thinking at scale, and increasingly the ability to evaluate model behavior in high-stakes domains.

It’s not sufficient to be a “prompt engineer” or an API wrapper developer — that tier is already being commoditized. The protected tier is researchers, ML engineers who understand the full training stack, and engineers who can build reliable inference infrastructure at scale.

3. Systems and Hardware Engineering

This is underappreciated in most “AI-proof jobs” lists because it lacks the glamour of AI research, but the structural protection is strong. Chip design, embedded systems, semiconductor engineering, and hardware architecture require deep physical intuition, multi-year development cycles, and the kind of accumulated expertise that doesn’t transfer easily to AI models trained on text.

Engineering roles that involve safety, infrastructure, and large-scale systems remain highly protected because errors have real-world consequences. The global semiconductor race — driven in large part by the demand for AI chips — is creating massive demand for hardware engineers that software automation genuinely cannot fill. Someone has to design the GPUs that run the models.

TSMC, NVIDIA, Intel, and AMD are all expanding headcount in hardware engineering even as they cut elsewhere.

4. Software Architecture and Systems Design

Routine coding is genuinely at risk — AI is already writing large amounts of production code, and entry-level software engineering jobs are contracting. But software architects design large-scale systems, and while AI can suggest patterns, it cannot fully understand enterprise-level complexity.

The distinction is between writing code (increasingly automatable) and deciding what to build, why, and how it fits into a complex sociotechnical system (not automatable). Senior architects who can reason about trade-offs, failure modes, organizational constraints, and 10-year system evolution are more valuable now than before AI, because they’re the ones who need to direct and validate the AI-generated code underneath them.

5. AI Governance, Ethics, and Regulatory Compliance

This is the fastest-growing new category and the one most people aren’t thinking about. The EU AI Act, US executive orders on AI, and emerging global regulatory frameworks are creating an entirely new profession: people who understand both the technical reality of AI systems and the legal, ethical, and organizational frameworks governing them.

Climate change, energy transition, and environmental regulation require long-term planning and policy decisions — roles that offer strong career longevity because they combine science, regulation, and ethics. The same logic applies to AI governance: these roles sit at the intersection of law, policy, engineering, and ethics, and they require human accountability in a way that cannot be delegated to the technology being regulated.

The honest caveat about “protection”

No tech field is fully immune — the more accurate framing is rate of displacement and which tier within a field is safe. Even in cybersecurity, routine vulnerability scanning and compliance reporting are being automated. Even in hardware engineering, some simulation and testing work is being automated.

The protection lies consistently at the senior, judgment-intensive, cross-disciplinary tier of any field — the people who make decisions under uncertainty, who are accountable for outcomes, and who can see the system as a whole rather than executing defined tasks within it

The bleakest finding from the research, worth being honest about

IBM has tripled entry-level hiring in 2026, saying that while AI can do many entry-level jobs, it still needs a human touch — but cutting entry-level jobs comes with the risk of erasing the pipeline needed to train future experienced workers and mid-level managers.

The protected senior roles still exist, but the traditional path to getting there — starting junior and climbing — is being disrupted. That’s the structural challenge that doesn’t have a clean answer yet.