AI Safety Researcher
You work on making advanced AI systems behave as intended — interpretability, alignment, evaluation.
Degree usually expected Hard, but with a more navigable on-ramp than general research, because the field has deliberately built one.
Can I actually do this?
Research-level, with one real difference from the other two: this field runs structured entry courses precisely because it wants more people in it. That lowers the discovery barrier, not the difficulty. Technical alignment work still assumes strong ML; governance-facing work assumes policy literacy instead.
Who it suits. People who want to work on failure modes rather than capabilities, and who can hold technical and normative questions at once.
Runway. Long, but with structured free entry courses that most research fields lack. We do not publish an hours figure, as the only one available is an unsourced third-party estimate.
Coming from research engineering or AI security? Both are genuine entries — technical alignment from one side, evaluation and red-teaming from the other. AI Research Engineer AI Security Engineer
Also advertised as
The route
Four stations, in order. Each one is a thing you finish before the next matters.
-
Station one
Learn it free
Only the best few, deliberately. Every one of these is free to use — the pill on each card says exactly what is and isn't free.
BlueDot Impact — AI Alignment course
Free to learn · no certificate
Structured alignment curriculum, published unit by unit and browsable without signing up. "Start for free" is their own wording — but note the cohort is APPLICATION-GATED rather than merely free: you apply, you do not simply enrol.
Verified 2026-07-26
BlueDot Impact — Technical AI Safety course
Free to learn · no certificate
Cohort-based technical safety course, free and application-gated; the curriculum is browsable without applying. TIME-SENSITIVE: when checked on 2026-07-26 the page read "Apply by 27 Jul", so that round has now closed or is closing. Check the live page for the current cohort rather than trusting this line.
Verified 2026-07-26
AI Alignment Forum
Free to learn · no certificate
Where much of the technical discussion happens in public, free to read, with curricula and sequences published openly.
Verified 2026-07-26
-
Station two
Attest strategically
Neither credential here is what gets you into safety research — the field hires on demonstrated work. IAPP AIGP is genuinely relevant to the GOVERNANCE side of this role, though we found no accreditation on its page; CertNexus CAIP is the accredited one but is general AI rather than safety. Do the free courses first: they are the actual on-ramp.
CertNexus Certified Artificial Intelligence Practitioner (CAIP)
CertNexus · CAIP
Free to learn · paid certificate Recognized
The honest cost Cost Not published
CertNexus does not print the exam fee on this page, rendered or not.Validity Not stated on the fetched pages. Renewal Not stated on the fetched pages. Assessment Certification exam, accredited to ISO/IEC 17024:2012 Proctored No Verify via unknown Cost per active year Unknowncannot be computed — no published fee Independently accredited by the ANSI National Accreditation Board — the only AI certification in our matrix that is.
Read this before you buyANAB/ISO 17024 accredits the CERTIFYING BODY's assessment process — a different and stronger claim than a vendor certifying competence on its own product. That is why it appears here rather than a vendor badge. Price, validity and renewal are all unpublished, so budget nothing until you see a checkout page.
Verified 2026-07-28 · Official page
IAPP Artificial Intelligence Governance Professional (AIGP)
IAPP · AIGP
Free to learn · paid certificate Recognized
The honest cost Cost Not published
IAPP does not print the fee on this page, rendered or not. Membership status typically affects certification pricing at IAPP, so check while signed in.Validity Not stated on the fetched page. Renewal Not stated on the fetched page. Assessment Certification exam; IAPP publishes a free Body of Knowledge and Exam Blueprint Proctored No Verify via unknown Cost per active year Unknowncannot be computed — no published fee Governance-facing credential covering AI law, risk and responsible-AI practice.
Read this before you buyNO independent accreditation was found on its page — a dated absence, not proof it will never have one. Compare CertNexus CAIP above, which is ANAB/ISO 17024 accredited. Both are vendor-neutral AI certifications; only one is independently accredited, and that difference is the point.
Verified 2026-07-28 · Official page
-
Station three
Prove it
A certificate says you passed a test. These say you can do the job.
An interpretability write-up
Take a small model, investigate a specific behaviour, and publish what you found — including the parts that stayed unclear.
An evaluation someone else can run
Build a reproducible evaluation for a specific failure mode. The field needs these and they are checkable.
A completed alignment curriculum with output
Finish one of the free courses above and publish the project it ends with. That project, not the course, is the evidence.
-
Station four
Get hired
Search these exact titles
Who hires for this. AI labs' safety teams, independent safety organisations, academic groups, and policy institutes.
This field is unusually open about its entry routes and published work is read by the people hiring, so publishing early is more normal here than in most research fields. That is our reading of how the field operates, not a hiring statistic we have verified.
On salaryWe don't publish salary estimates. Numbers copied between blogs drift from reality, and a wrong number costs you real negotiating power. When we have a verified public source, it goes here with its date.
Where this route continues
- AI Research Scientist — deeper into original research
- AI Research Scientist — sideways move
- AI Security Engineer — sideways move
Continues into senior safety research and AI policy.
This page last verified 2026-07-26 · How we verify
All three resources fetched and READ 2026-07-26. BlueDot's free-but-application-gated model and its dated deadline are quoted from their own pages, and the deadline is explicitly flagged as time-sensitive rather than presented as permanent.