Loading the job postings and counting them…
Loading the job postings and counting them…
AI Career Compass · role
Studies and reduces the risks of AI systems, and works on the rules around them. Alignment and safety research, red teaming, AI security, trust and safety, policy and governance.
Measured In our tracked sample, AI safety, security and policy roles make up 90 of 1,694 AI postings, or 5%.1
Measured The skills these postings ask for most are Python (asked for in 45 of 90), machine learning (43 of 90) and large language models (35 of 90).123
Written by an AI analyst (claude-opus-5-5) and checked by code against the numbers it cites.
What these postings say the work involves, counted from their lists of duties. A posting can list several.
Out of these 90 postings. “Asked for” adds required, one of several options, and listed; preferred is counted apart.
Programming, data, maths and statistics, and domain knowledge
| Skill | Asked for | Required | One of several | Listed | Preferred |
|---|---|---|---|---|---|
| Python | 45 of 90 | 9 | 9 | 27 | 3 |
| Software engineering practice | 23 of 90 | 7 | 0 | 16 | 8 |
| SQL | 14 of 90 | 6 | 1 | 7 | 2 |
| Go | 10 of 90 | 0 | 7 | 3 | 0 |
| Data analysis | 7 of 90 | 3 | 0 | 4 | 1 |
| Rust | 7 of 90 | 1 | 4 | 2 | 0 |
| C++ | 4 of 90 | 0 | 2 | 2 | 0 |
| TypeScript or JavaScript | 4 of 90 | 0 | 2 | 2 | 0 |
Modelling, training, retrieval, agents, evaluation and other methods
| Skill | Asked for | Required | One of several | Listed | Preferred |
|---|---|---|---|---|---|
| Machine learning | 43 of 90 | 8 | 1 | 34 | 9 |
| Large language models | 35 of 90 | 10 | 3 | 22 | 9 |
| Alignment and AI safety research | 25 of 90 | 3 | 0 | 22 | 4 |
| AI agents | 21 of 90 | 7 | 2 | 12 | 3 |
| Red teaming and adversarial testing | 10 of 90 | 2 | 0 | 8 | 8 |
| Fine-tuning and post-training | 10 of 90 | 1 | 0 | 9 | 2 |
| Evaluation (evals) | 8 of 90 | 2 | 1 | 5 | 4 |
| Deep learning | 5 of 90 | 0 | 0 | 5 | 2 |
Deployment, infrastructure, reliability, security, monitoring and cost
| Skill | Asked for | Required | One of several | Listed | Preferred |
|---|---|---|---|---|---|
| Security engineering | 33 of 90 | 10 | 0 | 23 | 4 |
| Cloud platforms (AWS, GCP, Azure) | 19 of 90 | 3 | 3 | 13 | 8 |
| Reliability and incident response | 14 of 90 | 3 | 0 | 11 | 2 |
| System design and architecture | 14 of 90 | 1 | 0 | 13 | 0 |
| Kubernetes and containers | 13 of 90 | 0 | 3 | 10 | 8 |
| MLOps and ML platforms | 13 of 90 | 3 | 1 | 9 | 5 |
| Monitoring and observability | 13 of 90 | 2 | 0 | 11 | 3 |
| Backend services and APIs | 13 of 90 | 2 | 1 | 10 | 1 |
Publishing and leading research
| Skill | Asked for | Required | One of several | Listed | Preferred |
|---|---|---|---|---|---|
| Open-source contributions | 3 of 90 | 0 | 0 | 3 | 6 |
| Leading research | 2 of 90 | 1 | 0 | 1 | 1 |
| Research publications | 0 of 90 | 0 | 0 | 0 | 4 |
Customers, communication, product thinking and leadership
| Skill | Asked for | Required | One of several | Listed | Preferred |
|---|---|---|---|---|---|
| Communication | 34 of 90 | 13 | 0 | 21 | 0 |
| Technical writing | 5 of 90 | 1 | 0 | 4 | 2 |
| Leading and mentoring | 4 of 90 | 1 | 0 | 3 | 1 |
| Working with customers | 2 of 90 | 0 | 0 | 2 | 2 |
| Product thinking | 1 of 90 | 0 | 0 | 1 | 0 |
AI rules, policy and trust and safety
| Skill | Asked for | Required | One of several | Listed | Preferred |
|---|---|---|---|---|---|
| Trust and safety | 14 of 90 | 4 | 0 | 10 | 6 |
| Risk and compliance | 10 of 90 | 3 | 0 | 7 | 5 |
| AI governance and regulation | 5 of 90 | 4 | 0 | 1 | 2 |
| Public policy | 3 of 90 | 1 | 0 | 2 | 2 |
Measured Knowledge of large language models is listed as preferred, meaning nice to have but not required, in 9 of the 90 postings we track.1
Measured Red teaming and adversarial testing, which means trying to make an AI system fail on purpose, is listed as preferred in 8 of the 90 postings we track.1
Employers list these as preferred. That they ask does not show that these skills decide who is hired.
The systems these postings expect people to build, run or work within.
Only what the postings state. Most do not say everything, so each number has its own count of postings.
47 of 90 postings state a minimum; the median is 5 years.
29 of 90 postings state a degree; 20 of them accept equivalent experience instead.
12 postings are remote. Of these, 12 name where you must live or work from, 0 do not say, and 0 say anywhere.
What employers advertise, grouped only when the ranges are comparable. Not a prediction of what you would earn.
Measured For full-time jobs in the US, the median of advertised range midpoints is 302,750 US dollars per year in base pay, based on 30 postings with comparable pay ranges.12
69 of 90 postings (77%) advertise a pay range. Employers that publish ranges may differ from those that do not, and some places require it by law, so these ranges do not describe all pay.
| Group of alike ranges | Postings | Median of range midpoints | Middle half of midpoints |
|---|---|---|---|
| USD, a year, base pay, United States, full-time10 employers | 30 | $302,750 | $270,250 to $355,500 |
| USD, a year, base pay, United States, employment type not stated2 employers | 16 | $362,500 | $305,625 to $417,500 |
This is advertised pay, not what anyone earns. We pool ranges only when currency, pay period, pay type, country and employment type are the same, and only with at least 10 of them. Nothing is converted. 23 other ranges could not be pooled (period not stated, several countries, or too few alike).
Advice built from what these 90 postings ask for. Tick what you can already do; we suggest up to three steps and one portfolio project, each with the requirement behind it and a way to show the result.
Analysis All 90 postings here come from a single day of data, 6 October 2026, when our tracking began, so we cannot show a trend yet.1
Measured These postings come from 28 employers that publish their jobs through public job feeds, so they are a sample, not the whole job market.1
Measured Many open listings are not new: only 24 of the 90 were first published in the 30 days up to our data cutoff, while 11 were first published over a year before it.12
How everything is counted: our method. All roles: AI Career Compass.