Research · AI security · Career · Layoffs
Is any job layoff-proof? What AI research says about AI security
The 2026 layoff numbers, what Anthropic and OpenAI found about which work AI changes first, four AI career areas that hold up, and a plain look at AI security as a long-term career.
Short answer
No job is layoff-proof, and the research doesn't claim one is. Anthropic's March 2026 study found no systematic rise in unemployment for AI-exposed workers so far, but slower hiring for people aged 22 to 25. OpenAI's April 2026 framework puts 18% of US jobs at relatively high automation risk and says work lasts where a person stays responsible for the result. AI security fits that description: someone has to answer for what an AI system does, and the US Bureau of Labor Statistics projects 21% growth for information security analysts from 2025 to 2035, against 3% for all occupations.
No job is layoff-proof. Anyone who sells you one is selling something. What you can do is move toward work that’s harder to cut, and in 2026 there is real research on what that work looks like.
This article goes through the layoff numbers, what Anthropic and OpenAI have published about AI and jobs, and where AI security sits in that picture. We sell an AI security course, so read the last part with that in mind. The research sections quote the sources and link to them.
The 2026 layoff numbers
Challenger, Gray & Christmas tracks announced US job cuts every month. Its report published on 5 November 2026 counted 617,189 cuts from January through October.
Employers cited AI as the reason for 121,401 of them, about 20%. Technology companies announced 171,501 cuts, 28% of the total and 21% more than in the same months of 2025.
US job cuts announced, January to October 2026
October itself was quiet. The month had 43,994 announced cuts, and only 1,265 were blamed on AI, the lowest monthly figure of the year. The firm’s read: “The reasons employers give have shifted from technology to the economy.”
Two cautions about these figures. They count what employers announce and the reason employers choose to give. A company can say “AI” when it means cost, and the reverse.
What Anthropic found
Anthropic’s economics team published “Labor market impacts of AI: A new measure and early evidence” on 5 March 2026. It combines what language models could do in theory with what people use Claude for at work, and calls the result observed exposure.
The headline finding is calmer than the news. The authors found no systematic increase in unemployment for highly exposed workers since late 2022.
The detail matters more:
- Computer programmers were the most exposed occupation, with 75% of their tasks covered. Data entry keyers were at 67%.
- 30% of workers had zero coverage. Cooks, motorcycle mechanics, lifeguards and bartenders are in that group.
- For workers aged 22 to 25, the rate of finding a job in the most exposed occupations fell by about 14% compared with 2022. The authors call that result “just barely” statistically significant and tentative. They saw no similar drop for people over 25.
Anthropic, March 2026: what the data shows so far
So the early signal is in hiring, and it’s the youngest workers who feel it.
Anthropic’s June 2026 Economic Index report adds a survey of Claude users. One in ten rated losing their own job as likely or very likely. More than a third said a junior colleague’s chance of losing their job was above 60%. Anthropic states plainly that this survey “is not representative of the general population”, so treat it as what heavy AI users expect.
What OpenAI found
OpenAI published “Modeling an AI jobs transition” on 25 April 2026. Its point is that “can AI do the tasks?” is the wrong single question. It asks three:
- Can AI perform a meaningful share of the occupation’s tasks?
- Is a person still central to delivering, supervising or taking responsibility for the work?
- If AI makes the service cheaper, will demand grow enough to absorb the gain?
Applied to 921 US occupations, that gives four groups:
OpenAI's four groups, share of US jobs
- 18% relatively high automation risk
- 24% likely to reorganize
- 12% could grow with AI
- 46% less immediate change
| Group | Share of jobs | Examples OpenAI gives |
|---|---|---|
| Relatively high automation risk | 18% | Data-entry clerks, telemarketers, proofreaders |
| Likely to reorganize | 24% | Lawyers, accountants, software developers, teachers |
| Could grow with AI | 12% | Tutors, mental-health counselors, financial advisers |
| Less immediate change | 46% | Electricians, plumbers, construction, food service |
OpenAI is careful about what this means: “These categories are not predictions that a particular percentage of jobs will disappear.” It also notes that effects may show up first in hiring, wages or the makeup of work, which matches what Anthropic measured.
What the two studies agree on
Read together, they point at the same things.
Work built on standardized information is the most exposed. If the job is moving text or numbers from one form to another by fixed rules, AI already does a lot of it.
Work lasts where a person has to answer for the outcome. OpenAI’s second question is the useful one for career planning. Ask it of your own role: if this goes wrong, does a named human have to explain why?
Entry-level roles are under the most pressure. Both studies raise it. A plan that depends on getting a junior job and learning there is riskier than it was three years ago.
Four AI career areas that hold up
These are our reading of the research, not a ranking from either lab. For each one we give the reason and the official US growth projection where the Bureau of Labor Statistics publishes one. BLS projects 3% growth for all occupations from 2025 to 2035, so compare against that.
Projected US job growth, 2025 to 2035
AI security. Securing the models, agents and data pipelines companies are deploying. BLS projects 21% growth for information security analysts, the closest category it tracks, and says the wider use of AI has increased the need for security. More on this below.
AI governance, risk and audit. Deciding which AI uses are allowed, documenting them and proving it to a regulator or customer. This is accountability work by definition. Certification bodies have built exams for it, such as IAPP’s AIGP and ISACA’s AAIA.
Building and running AI systems. OpenAI puts software developers in the “reorganize” group: AI writes more of the code, and the person carries the design and the responsibility. BLS still projects 10% growth for software developers, quality assurance analysts and testers. The developers who learn to build with models, evaluate them and operate them are on the right side of that reorganization.
Data science and evaluation. Someone has to measure whether an AI system works, on real data, before and after it ships. BLS projects 35% growth for data scientists.
One honest note: OpenAI’s “less immediate change” group is mostly physical, place-based work. If you want distance from AI more than you want a job in AI, an electrician’s license is a serious answer.
Why AI security is a long-term career
Three reasons, each tied to something in the research.
Somebody has to be accountable. When an AI agent sends customer data to the wrong place, the company can’t blame the model. A person owns that risk. That is OpenAI’s second question, answered yes.
Every AI deployment adds work for security. The same adoption that reorganizes other jobs creates new systems to protect. OWASP moved Excessive Agency, an agent with more permissions than it needs, up to third on its 2026 list of LLM risks. That only happens when companies ship agents faster than they secure them.
The work doesn’t stay standardized. Attackers change methods when a defense works. A job where the other side adapts is hard to reduce to fixed rules, which is the opposite of the profile both studies flag as exposed.
The institutions are moving too. ISC2 asked its members in July 2026 to help define an AI security credential, and SANS, CompTIA and ISACA already have exams. Training bodies build certifications where employers are asking for skills.
The limits, stated plainly
AI security doesn’t make you immune.
Security teams get cut in bad years like everyone else. Technology companies announced 171,501 cuts this year, and security teams work inside those companies.
AI will take over parts of security work, starting with repetitive alert triage. The people who last will be the ones who direct those tools and answer for the results.
The entry-level squeeze applies here as well. A course and a certificate won’t get you hired on their own. Proof of work does: labs you ran, a write-up of an attack you reproduced, a project on GitHub.
And the field is young. Job titles and required skills are still settling.
How to start, without quitting your job
Most people move into AI security from a role they already have: cloud, platform, development, SOC, IT or governance. You keep the job and add the skill.
- Read one page on the subject: LLM security: the risks, the fixes and where to start.
- Follow the order in How to become an AI security engineer.
- Do the free labs on your own laptop. Ours are public on GitHub, and PortSwigger has free LLM attack labs too.
- Decide whether you need an exam: AI security certifications in 2026.
If you want a structured path, compare the options in Best AI security courses for 2027. One of them is ours, AI Security Engineer: practical AI security, not a theory course, with a free account so you can try lessons before paying.
Sources
Checked on 14 November 2026.
- Challenger, Gray & Christmas, October 2026 job cuts report, 5 November 2026.
- Anthropic, Labor market impacts of AI: A new measure and early evidence, 5 March 2026.
- Anthropic, Economic Index report: Cadences, 26 June 2026.
- OpenAI, Modeling an AI jobs transition, 25 April 2026.
- US Bureau of Labor Statistics, Occupational Outlook Handbook, 2025 to 2035 projections: information security analysts, data scientists, software developers, quality assurance analysts and testers.
- OWASP GenAI Security Project, Top 10 for LLM Applications 2026.
- ISC2, AI security certification development, July 2026.