AI safety roles up 91% as hiring freezes: analysis

The job board is not dead; it is pickier. One tab shows broad hiring caution, while another asks for people who can make AI systems safer, more governable, and less likely to create risk at product speed. That does not make AI safety an easy lane, but it does make it one of the more useful places to study the difference between AI hype and actual workflow.

Business Insider reported that LinkedIn research found AI safety roles rose 91% over the past year. At LinkedIn's Talent Connect conference, CEO Daniel Shapero said hiring has largely "frozen," while also calling AI safety one of the growth pockets. The career lesson is not that every learner should become a model researcher. It is that the AI labor market is splitting into narrower jobs where risk, policy, security, and product judgment matter.

The growth pocket is real, but the label is broad

Business Insider described AI safety as work focused on making sure AI systems operate safely and ethically, with roles spanning technical research, risk mitigation, policy work, and the integration of safety frameworks into AI products. That is a wide umbrella, which is where applicants can get into trouble. An AI safety posting may sit closer to machine learning evaluation, trust and safety operations, governance, child safety policy, or product launch review. Treat the title as a clue, not a definition.

@title AI safety work branches
@source Business Insider

  AI safety
     │
     ├─ Technical research
     ├─ Risk mitigation
     ├─ Policy work
     └─ Safety frameworks

@caption AI safety roles split across research, risk, policy, and product work.

That branchiness is good news for people who are AI adjacent but not trying to become research scientists. It also means credential inflation will show up fast: a generic AI certificate may help with vocabulary, but it will not prove you can reason through misuse, policy, safeguards, or product tradeoffs. If you are choosing a course, favor one that leaves you with a concrete artifact: a risk register, an evaluation plan, a model use policy, or a safety review memo. Those are easier to explain than another badge that says AI fundamentals.

The wider AI hiring picture is uneven

Indeed Hiring Lab's January 2026 US labor market update found hiring activity remained subdued, with job postings flat or declining in many occupations. The same report said jobs with AI mentions were growing across many knowledge work occupations, and Indeed's AI Tracker reached 4.2% in December 2025. It also found nearly 45% of data and analytics postings contained AI related terms, compared with about 15% in marketing and 9% in human resources. That spread matters because AI literacy is not landing evenly across functions.

Indeed Hiring Lab's March 2024 update gives useful context for why a single hot phrase can mislead job seekers. It said AI related postings had risen to 2% of all jobs at the end of February 2024, but remained below their March 2022 peak of 3.3%. In other words, AI hiring can rebound inside a weak market without becoming a universal escape hatch. The practical move is to compare roles by workflow and function, not by whether the job title contains AI.

Pay is a signal, not a shortcut

Business Insider reported that data scientists on OpenAI's Safety team can earn up to $325,000 a year, while Anthropic listed compensation of up to $285,000 for a safeguards enforcement analyst focused on child safety. The same report said Google was hiring a policy strategist for kids and families paying up to $239,000, plus a 20% target bonus and equity, and that a program manager on Meta's Integrity GRC team could earn up to $209,000, in addition to bonus and equity. Those ceilings are real signals that companies value the work. They are not proof that every AI safety role pays at that level, or that a short course will get you there.

For a 25 year old, the constraint may be proof of judgment without much work history. For a 45 year old, it may be translating experience in compliance, security, product, education, healthcare, or policy into AI risk language without pretending to be an ML engineer. The OECD frames AI governance around harnessing benefits while mitigating risks, and that is a useful way to position the work. Safety careers are not only about saying no to AI; they are about making adoption legible, tested, and accountable.

What to build next

Business Insider's reporting on LinkedIn's 91% growth figure should put AI safety on the shortlist for learners who want an AI adjacent path with clearer business need than buzzword chasing. Start by deciding which branch you are actually targeting: technical research, risk mitigation, policy, or product safety frameworks. Then build one portfolio piece that matches that branch, rather than collecting broad AI credentials with no work sample attached.

The next thing to watch is whether AI safety stays concentrated at major AI companies or spreads deeper into ordinary employers adopting AI tools. Indeed Hiring Lab's data already shows AI terms appearing across knowledge work, but the useful jobs will be the ones with defined responsibilities, not decorative titles. If the market remains selective, clarity becomes a career advantage: know the risks you can evaluate, the documents you can produce, and the teams you can help.