EXPOSED verdict contested
This is a residual category — model validators, quantitative analysts, cryptologic and biostatistical specialists who don't fit the statistician, actuary, or data-scientist codes — and the modal worker spends the day writing and testing code, cleaning data, fitting and documenting models, and producing technical memos. Those are exactly the artifacts LLMs now draft at usable quality, and nothing about the work is physical or licensed. What holds is the accountable end: choosing formulations under ambiguity, deciding whether a model is fit for a consequential decision, and signing the validation report someone in risk or regulatory will lean on.
Most of this decline happened after 2021 — it is not the pandemic dip.
This line is counted by the Bureau of Labor Statistics — the one figure on this page that isn't a judgement of ours. Headcount moves on demand, offshoring, demographics and the business cycle, and automation is one term among several, often not the loudest.
So a falling line is not evidence that AI did it, and a rising one is not evidence that it won't. Both happen in this register: some occupations resist automation and shrink anyway, others are highly automatable and keep growing.
BLS projection, 2024–2034
+4%
Percentage only. The projection counts a different population from the 3,720 above — it includes self-employed workers, which for this occupation is most of them, so the two headcounts are not comparable.
Exposed, but growing
AI can already do a lot of these tasks, and the BLS still expects +4% more of these jobs by 2034. Demand for the output is growing faster than the work is being automated away — the mechanism BLS gives for software developers, and the combination people most often misread as an error.
Different clocks. The score is what current AI could do to this work today. The projection is how many of these jobs will exist in 2034. Everything between the two — how fast employers actually adopt, whether demand grows in the meantime — is why they can point opposite ways without either being wrong.
~300 openings a year on average, including replacing people who leave.
GeometerCryptanalystCryptologistData AnalystGeometricianBiotechnicianWeight AnalystData TechnicianBalance EngineerHarmonic AnalystResearch AnalystResearch ScientistSecret Code ExpertDatabase TechnicianLog Data TechnicianBiometrics TechnicianField Data TechnicianSubmission TechnicianBioinformatics AnalystMathematical ScientistMathematics TechnicianBioinformatics SpecialistBioinformatics TechnicianMuseum Informatics Specialist
The BLS uses Mathematical Science Occupations, All Other for work that doesn't fit any named occupation, so it covers roles that have little in common with each other. Two consequences worth knowing before you read anything below:
Holding it up: judgment & accountability . Weakest point: liability shield .
Mixed — a routine tier and a judgment tier At 9 the split is visible inside a single day: the derivation setup, the choice of estimator for a badly-specified problem, and the argument for why a stress scenario is binding still take a person, but the Python that implements it, the backtest harness, the sensitivity tables and the SR 11-7 style write-up are now first-drafted by tooling faster than you can type them.
Fully desk- and screen-based A 3 reflects that the only physical constraints are a workstation and, for the cryptologic and defence-adjacent slice of this code, a SCIF or air-gapped terminal you must physically sit at — that limits where the work happens, not what the hands do.
No licence, no signature requirement There is no licence to strip: unlike the actuary who signs a Statement of Actuarial Opinion under ASOP, a model validator's report is absorbed by the CRO or model risk committee, and an SOA/CFA credential or clearance is a hiring filter rather than a legal barrier to anyone else doing the work.
Meaningful discretion 12 is the effective-approve call: deciding a model is unfit for a capital decision or that a proxy variable invalidates a published estimate is a discretionary judgment with real consequences, but it lands inside a governance chain with committee sign-off, documented methodology standards, and an escalation path above you rather than terminating with your name.
The verdict above describes this occupation as a whole. Almost nobody does the typical version of a job — tick what's actually in your week and see how your own mix sits.
Your task mix speaks to task resistance (9/20 here) — how much of the day's work current AI already does. That is the dimension the boxes above are about.
It cannot move the other three. Liability shield (3/20) is whether the law requires a licensed human to sign. Trust premium (7/20) is whether buyers specifically pay for a person. Judgment and accountability (12/20) is whether the role exists to own consequential calls. Those are facts about the occupation's standing, not about which tasks are in your week — a paralegal who does only trial exhibits still holds no licence. Together they are 22 of this occupation's 34 points (65%).
Embodiment (3/20) is also a property of the work rather than the worker, but we don't tag individual tasks as physical or not, so the picker can't tell you anything about it. That's a limit of this tool, not a claim.
Did we get the list right? Tell us what's missing — the tasks are written from the outside, and you're reading this from the inside.
The moves above are yours to make. This is the other half: what would have to change in the world for the occupation itself to score higher. None of it is in any one person's gift, but it is where the floor actually comes from. Scores here are not a one-way ratchet. Only two of the five dimensions — task resistance and embodiment — track what machines can do. The other three track law, what buyers will pay for, and who is answerable, and those move in both directions, often in response to the same pressure AI creates. If every lever below landed, this occupation would score around 50/100, still EXPOSED.
Independent validation of AI/ML models becomes a mandated second line function under EU AI Act conformity assessment and NAIC/state insurance model-governance bulletins, so the role owns the go/no-go on deployment rather than advising on it — the validator's adverse finding blocks release.
Task-mix shift: once code generation, data cleaning and memo drafting are absorbed, the residual day is formulation choice under ambiguity, identification/assumption critique, adversarial testing of generated models, and defending findings to regulators and examiners — genuinely two-tiered work where the upper tier is what remains. Also rises if validating AI systems built by others becomes the dominant workload.
Model risk management rules move from 'effective challenge by qualified staff' to a named-individual attestation: e.g. if the Fed/OCC/FDIC SR 11-7 successor or an EU AI Act Article 17 implementing act requires a specifically identified model validator to sign that a high-risk model was independently validated, with personal accountability under a regime like the UK SM&CR / FCA Senior Manager attestations already applied to some model owners. Similar route via FDA biostatistics: a named responsible statistician signing the SAP and unblinding memo under ICH E9(R1).
Narrow route only: cryptologic and classified-clearance work, plus expert-witness and regulatory-examination testimony, where the buyer needs a cleared or deposable human. Confined to a minority of this residual code.
The limit. Embodiment has no route. The liability route is the only large one, and it is uncertain because model risk regulation has historically attached to the institution, not to a licensed individual — there is no board, no license, no personal statutory duty to build on, so any attestation regime would be contractual or supervisory rather than professional licensure. Realistic ceiling in the low-to-mid 50s, and that requires the named-signer rule to actually land.
| Chicago-Naperville-Elgin, IL-IN | 1,100 | $82,230 +1% |
| Seattle-Tacoma-Bellevue, WA | 240 | $72,710 -11% |
| Birmingham, AL | 80 | $62,630 -23% |
| New York-Newark-Jersey City, NY-NJ | 70 | $79,460 -2% |
| Oklahoma City, OK | 40 | $36,820 -55% |
| Phoenix-Mesa-Chandler, AZ | 40 | — |
| Atlanta-Sandy Springs-Roswell, GA | 30 | $93,320 +15% |
| Austin-Round Rock-San Marcos, TX | 30 | $97,190 +19% |
| Austin-Round Rock-San Marcos, TX | 30 | $97,190 +19% |
| Atlanta-Sandy Springs-Roswell, GA | 30 | $93,320 +15% |
| Chicago-Naperville-Elgin, IL-IN | 1,100 | $82,230 +1% |
We have no reported case of a named organisation automating this occupation. Not one deployment, not one announcement.
That is worth saying out loud next to a score of 34. The verdict above is about what the work exposes — what current AI could do to these tasks. It is not a claim that anyone has done it. For this occupation those two things have come apart completely: the capability argument is on this page, and the evidence column is empty.
Has AI actually changed your work? One tap, anonymous, and the running tally is public. Nothing else is asked of you.
Rather than check back: get the digest and we'll tell you what changed — or watch a single occupation from its own page.