← Risk register SOC 15-2021 · reviewed 2026-08-11

Mathematicians

2,030 US workers · median $126,710/yr · Tech

COOKED

Almost everything a mathematician does — deriving proofs, building and validating models, running symbolic and numerical analysis, writing up results — is symbol manipulation on a screen, which is precisely where frontier models have improved fastest (olympiad-level proof, formal verification via Lean, automated model fitting). There is no license, no stamp, no physical component, and the buyer is usually an institution paying for results rather than a person paying for a relationship. What survives is at the top of the field: choosing which problems matter, formulating fuzzy real-world questions as tractable mathematics, and vouching for whether a machine-generated proof or model is actually correct and applicable.

10-year outlook: Headcount in this small occupation stays flat or shrinks as AI absorbs derivation and analysis work, and the remaining jobs concentrate in problem formulation, verification, and security-cleared or regulated modeling roles.

US employment, 2019–2025-22.8%
2,6302,030 workers

Nearly all of this fall was the 2020 shock. It has been climbing back since.

Median pay $105,030 → $126,710 -3.5% in real terms (nominal +20.6%, less ~25% US inflation over the period)

The job count is not the verdict

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. The marked year is 2020.

BLS projection, 2024–2034

-0.7%

Percentage only. The projection counts a different population from the 2,030 above — it includes self-employed workers, which for this occupation is most of them, so the two headcounts are not comparable.

Exposed, and shrinking

Both signals point the same way: the tasks are largely automatable and the BLS projects -0.7% by 2034. This is the case where the score and the forecast agree, and it is the one worth taking seriously.

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.

~100 openings a year on average, including replacing people who leave.

One email if this score changes. Watch as many occupations as you like from the same address — no account, and nothing is sent on a schedule, only when a verdict actually moves.

Also known as — 17 job titles this covers

Titles reported by people doing this work, from the US Department of Labor's O*NET survey. If your job title is here, this page is about your work even though the name doesn't match.

AlgebraistResearcherCryptanalystGeometricianCipher ExpertCryptographerMathematicianGame MathematicianResearch ScientistAgent-Based ModelerApplied MathematicianComputational ScientistEngineering MathematicianComputational MathematicianResearch Computing SpecialistCryptographic Vulnerability AnalystMath Researcher (Mathematics Researcher)

Score — 30/100 resistance

Holding it up: judgment & accountability (12/20). Weakest point: liability shield (1/20).

Five dimensions, 0–20 each, summed. Higher means more protected. The arithmetic is shown so you can check it: 8 + 2 + 1 + 7 + 12 = 30. · Scored 2026-08-11, and re-examined when evidence accumulates rather than on a schedule.

Task resistance 8/20

Mixed — a routine tier and a judgment tier Proof search, symbolic manipulation, numerical simulation, statistical model fitting and LaTeX write-up are all now partly machine-executable — Lean-verified lemma chains and automated fitting cover a real share of the working day — but the step of turning a defense contractor's or actuarial client's vague question into a well-posed theorem or model specification still has no reliable automated equivalent, which is what keeps this at 8 rather than in the 0-6 band.

Embodiment 2/20

Fully desk- and screen-based The entire job runs on a laptop, a compute cluster, and a whiteboard; the only physical requirements are showing up to a seminar or conference room, which is why this sits at 2 and not 0.

Liability shield 1/20

No licence, no signature requirement There is no mathematics license, no PE-style stamp, no board that can strike you off — an incorrect proof or a flawed model gets you a retraction or a lost contract, not a personal cause of action, so the 1 reflects reputational exposure only.

Trust premium 7/20

Some relationship component Grant panels, cryptographic reviewers and internal clients do care whose name is on a proof — a Fields-adjacent reputation buys benefit of the doubt on a 200-page argument — but the deliverable is still the theorem or the model, which is checkable independently of who produced it, putting this at 7 rather than in physician-or-therapist territory.

Judgment & accountability 12/20

Meaningful discretion Choosing the abstraction, deciding which assumptions to relax, and judging whether a machine-generated proof step is actually sound or a plausible-looking hallucination are genuinely unbounded calls with consequences that show up years later in cryptographic standards or risk models — that is real discretion, but you are typically one contributor inside a peer-review or team structure rather than the single signatory, which caps it at 12.

This occupation has already been through one. Headcount fell 25.6% between 2017 and 2025 — 2,730 to 2,030 — while the median wage held roughly flat in real terms ( -5.7% after inflation). A job being commoditised usually loses pay along with headcount. One that shrinks by half while pay holds is leaving a specialist core behind, and the score above was assigned from the occupation title without sight of that history — so it may be describing the job this used to be rather than the people still doing it. Why this is a known limit.

Confidence: medium · reviewed 2026-08-11 · how scoring works

What this job involves — and which parts are yours

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.

AI already does these at usable quality

These still need a person

Active moats on the surviving side: judgment

How to future-proof this job

Training paths for your skill gaps: CS50x, Harvard — how software is actually built free · Coursera — quality control and inspection courses, auditable free free to audit · Khan Academy — physics, chemistry and biology from the ground up free · Coursera — people management and team leadership specialisations free to audit

All 35 skills ranked by how many jobs they open →

Where this experience transfers — occupations you could move toward

Computed from U.S. Dept. of Labor O*NET skill and knowledge profiles: high overlap with what you already do, a materially higher resistance score, no large jump in required training, and no licence you would have to start a new pipeline to get. Targets that pay meaningfully less, that are themselves COOKED, or whose own headcount is falling are excluded — a move into a shrinking trade is not an escape.

Physicists EXPOSED · 46/100 · you already have ~69% of the skill profile

Skills to close: Technology Design, Quality Control Analysis, Science, Management of Personnel Resources

What would move this back up — beyond any one person

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 43/100 — EXPOSED.

4 specific changes that would raise this score
  • already happening task resistance +4

    Task-mix shift is genuine here: the field has a routine tier (lemma-chasing, symbolic manipulation, standard model fitting, literature-checking) and a judgment tier (problem selection, formulating messy domain questions as tractable mathematics, deciding whether a formal proof's hypotheses actually match the real system). If the routine tier is fully absorbed, what remains is measured almost entirely as judgment work, and a day's residual becomes harder to automate — visible already in how Lean/mathlib formalization projects (Terence Tao's PFR and Equational Theories projects) reallocate human effort toward statement design and blueprint decomposition rather than proof steps.

  • plausible judgment accountability +4

    If model-risk governance regimes that already name a responsible modeler extend to AI-generated mathematics: Federal Reserve/OCC SR 11-7 model validation requires a named independent validator to attest that a model is conceptually sound, and EU AI Act Article 14 human-oversight duties for high-risk systems push the same way. A bank, insurer, or actuarial employer designating a specific quantitative person as the accountable validator of AI-derived models makes ambiguity-ownership formal rather than cultural.

  • plausible liability shield +3

    Mathematicians as such have no license, but adjacent credentialed routes exist and can be pulled inward: the Society of Actuaries/CAS credentials and NAIC Actuarial Opinion requirements already require a named qualified actuary to sign reserve opinions, and CFA/FRM-style model attestation is increasingly written into bank policy. If a regulator or standards body requires a credentialed signatory on safety-critical or financial models regardless of whether AI produced them, the pool of mathematical work behind a personal signature grows. Also watch cryptographic standards: NIST post-quantum processes require named human review of security proofs.

  • plausible trust premium +2

    Narrow but real: mathematical expert-witness testimony (statistical evidence, patent damages, redistricting) requires a human who can be cross-examined, and Federal Rule of Evidence 702 plus Daubert gatekeeping mean courts admit a person's opinion, not a system's output. If courts and arbitration panels keep excluding non-attributable machine analysis, the litigation-consulting slice of the occupation carries a durable human premium — though this covers a small fraction of the 2,030 workers.

The limit. Even with every lever above, this occupation stays in the lower band. The core reason is structural and not fixable by regulation: the work product is symbol manipulation whose correctness is machine-checkable, so the usual argument for a mandatory human signature — that only a human can be held to account for an unverifiable judgment — is weakest exactly here. Formal verification makes proofs MORE auditable without a person, not less. The occupation is also tiny (2,030), which means no professional body has the mass to lobby a licensure regime into existence, and most people doing mathematics for pay are classified under other SOC codes (actuaries, statisticians, operations research analysts) where the shields already sit.

These are conditions, not forecasts — what would have to happen, not what will. Specific rules, cases and bills are named so you can go and check whether they exist and where they stand; verify before relying on any of them. Nothing here is legal or financial advice.

Where this work is, and what it pays there

BLS metro figures for 8 areas. The verdict above does not change by city — the rubric judges what the work involves, not where it happens — but pay and headcount do, and the national median hides a very wide range.

Most of these jobs

Washington-Arlington-Alexandria, DC-VA-MD-WV 200 $164,470 +30%
New York-Newark-Jersey City, NY-NJ 150 $113,590 -10%
Providence-Warwick, RI-MA 100 $109,260 -14%
Seattle-Tacoma-Bellevue, WA 70 $185,520 +46%
Chicago-Naperville-Elgin, IL-IN 60 $130,970 +3%
Boston-Cambridge-Newton, MA-NH 40 $104,190 -18%
Las Vegas-Henderson-North Las Vegas, NV 40 $102,890 -19%
San Diego-Chula Vista-Carlsbad, CA 40 $156,480 +23%

Best paid

Seattle-Tacoma-Bellevue, WA 70 $185,520 +46%
Washington-Arlington-Alexandria, DC-VA-MD-WV 200 $164,470 +30%
San Diego-Chula Vista-Carlsbad, CA 40 $156,480 +23%

Percentages are against this occupation's national median of $126,710. Counts are jobs in that metro, not vacancies. Metros where the BLS suppressed the cell are absent rather than shown as zero.

Who is actually doing this — nobody, on the record

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 30. 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.

Read that as a gap in the reporting we can see, not proof of absence — the dispatch runs on English-language feeds and misses plenty. If you know of a case, tell us, or add a field report from inside the job.

Quick take — do you do this job?

Has AI actually changed your work? One tap, anonymous, and the running tally is public. Nothing else is asked of you.

Self-reported and unverified — a sentiment signal, not a survey. One response per person per occupation; you can change your answer.

Field reports — what people say has changed

No field reports yet. A written account takes a paragraph rather than a tap, goes to an editor before it appears, and is the one thing on this page the rubric cannot produce on its own.

File a field report

Concrete beats general: a tool that arrived, a task that moved, a headcount decision you watched happen. Don't include anything that identifies you or your employer if that would put you at risk.

Watch this verdict
Kept current

Rather than check back: get the digest and we'll tell you what changed — or watch a single occupation from its own page.