AI Ethics Researcher Job Description Templates for Small AI Teams
6 templates for companies shipping AI without a research division: applied ethics researcher, algorithmic fairness, policy and governance, safety and red team, split role, and fractional advisor. Download as DOCX.
We ship an AI onboarding wizard inside FirstHR. The week it went live, a customer asked me something I could not answer in one sentence: what happens if the model handles two new hires differently for a reason nobody wrote down. That question is the entire job of an AI ethics researcher, and most job descriptions for the role never get near it.
The postings you find online were written at large research labs. They ask for a doctorate, publications at peer-reviewed venues, and experience running cross-functional programs, because that is who those organizations hire. Copy that into a posting for a company of twenty and you filter out every person who could actually do the work you have.
The six templates below are written for the other side of the market: companies with models in production, real regulatory exposure, and no research division. They cover applied ethics research, algorithmic fairness, policy and governance, safety and red teaming, an honest split role, and a fractional advisory engagement.
TL;DR
An AI ethics researcher measures and documents the harms an AI system can cause, then turns findings into changes engineering ships. No BLS occupation matches the title. Nearest benchmarks run from $80,730 for compliance officers to $140,300 for computer and information research scientists. Six templates below, downloadable as DOCX.
What the Role Actually Covers
An AI ethics researcher identifies, measures, and documents the harms an AI system can cause, then converts those findings into changes the engineering team ships. In practice the title covers four fairly different jobs, and the candidate pools for them barely overlap.
The confusion is expensive. A company that needs slice-level evaluation of a production model hires a philosopher, or a company facing a procurement questionnaire hires a statistician. Both people are good. Neither is doing the work that was needed. Decide which quadrant you are buying before writing a word of the posting.
Measurement and evaluation
The quantitative half
Building the tests that show how a model behaves across groups and edge cases, choosing the metric, and setting the number at which a gap becomes a blocker. This half needs statistics and direct access to your data, not a philosophy seminar.
Policy and governance
The written record
Tracking which rules reach your product, converting them into internal requirements, and producing the documentation procurement and auditors ask for. Closer to compliance than to research, and often the half a small company needs first.
Safety and red teaming
Adversarial testing
Deliberately breaking your own system: prompt injection, jailbreaks, data leakage, unsafe automation in agentic features. Overlaps heavily with security research, and the strongest candidates often come from there rather than from ethics.
Academic research
Rarely what you need
Publishing novel methods at peer-reviewed venues. This is what most copied job descriptions describe, and it is the wrong ask for a company with one model in production and no research budget. Do not require a publication record you cannot use.
Pick the Quadrant Before You Pick the Title
Ask one question: what is the first deliverable? If it is a number showing how the model behaves across groups, hire the measurement half. If it is a document a customer or auditor will read, hire the policy half. If it is a list of ways the system can be made to misbehave, hire from security research. If it is a paper, you are a research lab and this page is not written for you. Companies that cannot answer the question tend to write a posting asking for all four and hire nobody.
What Belongs in the Posting
A job description for this role does four things at once: it describes real work, it filters a pool that skews heavily toward the wrong background, it protects you legally, and it convinces a senior candidate that the authority is real. Most postings only manage the first.
The parts candidates read first
Which systems they will actually evaluate
Who they report to and whether that person can stop a launch
Whether the role is measurement, policy, safety, or all three
Team size and whether they are the first hire in this function
The parts that filter applicants
Degree stated as required or preferred, not implied
Whether hands-on analysis is expected or advisory only
Publication expectations, if any actually exist
Data access: their own queries or a request queue
The parts that protect you
FLSA classification stated on the posting
Essential functions written plainly
Confidentiality and disclosure expectations
Equal opportunity statement
The parts that win the hire
A pay range, which this candidate pool expects
Real authority: a named review gate, not a suggestion box
What happens when they say no to a launch
Support for distressing content review, where relevant
The omission that costs the most is authority. Experienced candidates screen for it in the first interview, and a posting that describes influence without a mechanism reads as decoration. Our guide to writing a job description covers the general structure, but for this role the authority section is the part that decides whether strong people apply.
6 AI Ethics Researcher Job Description Templates to Download
Download all six as one file or copy them individually. Each follows the same structure: company overview, position summary, key responsibilities, required qualifications, a classification and compliance note, an equal opportunity statement, and how to apply. The bracketed fields are the only parts you need to change.
Download All 6 AI Ethics Researcher Job Description Templates
Applied researcher, algorithmic fairness, policy and governance, safety and red team, split role, and fractional advisor. All in one download.
AI Ethics Researcher
In-house, applied
The general first-hire version: evaluation, documentation, pre-launch review, and the risk register, written for a company with models already in production.
Algorithmic Fairness Researcher
Quantitative evaluation
For the measurement half: slice-level testing, metric selection, the demographic data problem, and mitigations that survive the next model update.
AI Policy and Governance
Rules and documentation
For regulated customers and procurement questionnaires: jurisdiction mapping, control requirements, review gates, and board-level briefings.
AI Safety and Red Team
Adversarial testing
For generative and agentic features: attack the system, turn findings into regression tests, and set release criteria that hold under pressure.
Responsible AI Lead
Honest split role
For teams too small for a dedicated hire, with the percentage split written into the job and a named executive required to change it.
AI Ethics Advisor
Fractional or contract
For a scoped engagement before a launch or an audit, with deliverables, commercial terms, and a contractor classification note that takes the risk seriously.
Template 1: AI Ethics Researcher, In-House and Applied
The general first-hire version, written for a company that already has models in production: evaluation, documentation, pre-launch review, and an owned risk register.
AI Ethics Researcher Job Description (In-House, Applied)
AI ETHICS RESEARCHER JOB DESCRIPTION (IN-HOUSE, APPLIED)
Company: __ ([City, State] / Remote)
Reports to: [CTO / Head of Product / General Counsel / Founder]
Employment type: Full-time
FLSA status: Exempt, learned professional (see classification note)
Compensation: $_ to $_ per year
ABOUT [COMPANY NAME]
[Company Name] builds [product description] used by [customer type]. Our
[feature or model] makes or supports decisions that affect [users, applicants,
patients, borrowers, employees]. We are hiring our first AI Ethics Researcher to
tell us where those decisions go wrong before our customers do.
POSITION SUMMARY
The AI Ethics Researcher identifies, measures, and documents the harms our AI
systems can cause, translates those findings into changes engineering will
actually ship, and produces the written record that customers, auditors, and
regulators ask for.
KEY RESPONSIBILITIES
•Run structured evaluations of [models or features] for bias, accuracy gaps
across groups, privacy exposure, and failure modes under realistic use
•Design the evaluation itself: define the harm, choose the metric, build or
source the test set, and state what result would count as unacceptable
•Write model cards, system cards, and data documentation for [each release]
•Review new features at the design stage and file blocking concerns early
enough that they can be addressed
•Maintain our AI risk register: what could go wrong, how likely, how bad, what
mitigation is in place, who owns it
•Translate external frameworks and applicable law into internal requirements
engineers can implement
•Answer customer security and AI governance questionnaires with the technical
detail they ask for
•Brief [leadership / the board / the customer advisory group] on findings on a
[monthly / quarterly] cycle
REQUIRED QUALIFICATIONS
•[PhD / master's degree] in computer science, statistics, philosophy, law,
public policy, sociology, or a related field, or equivalent applied experience
•[Number] years of work on fairness, safety, privacy, or governance of
machine learning systems
•Ability to run your own analysis: [Python, SQL, statistical testing, notebook
workflows] without waiting for an engineer
•Written communication strong enough that a memo changes a roadmap
•Comfort delivering findings that are unwelcome, to people who outrank you
NICE TO HAVE
•Experience with [NIST AI Risk Management Framework / ISO 42001 / internal
audit / model risk management]
•Domain depth in [our vertical]
•Publication or open-source evaluation work we can read
CLASSIFICATION AND COMPLIANCE NOTE (read before posting)
A full-time researcher whose primary duty is advanced analysis requiring
knowledge customarily acquired through a prolonged course of specialized
intellectual instruction is generally an exempt learned professional under the
FLSA, provided the salary tests are met. The title alone does not create the
exemption: routine data labeling, test execution, or report formatting is not
advanced knowledge work, and a role built mostly from those tasks is non-exempt
and owed overtime past 40 hours in a week. Confirm your state salary threshold,
which may sit above the federal one. This is general information, not legal
advice.
EEO STATEMENT
[Company Name] is an equal opportunity employer and provides reasonable
accommodations for the essential functions of this role.
COMPENSATION AND HOW TO APPLY
Compensation: $_ to $_ per year, [equity], [benefits summary]
To apply, email __ with your resume and one example of an
evaluation you designed, including what you decided the threshold should be.
Template 2: Algorithmic Fairness and Bias Evaluation Researcher
For the measurement half of the role: slice-level testing, metric selection, the demographic data problem, and mitigations that survive the next model update. Pair it with the data scientist templates if the person will sit inside an analytics team.
Algorithmic Fairness and Bias Evaluation Researcher Job Description
ALGORITHMIC FAIRNESS AND BIAS EVALUATION RESEARCHER JOB DESCRIPTION
Company: __ ([City, State] / Remote)
Reports to: [Head of ML / Head of Data Science / CTO]
Employment type: Full-time
FLSA status: Exempt, learned professional (see classification note)
Compensation: $_ to $_ per year
ABOUT THIS ROLE
[Company Name] operates [models] that influence [hiring, lending, pricing,
triage, moderation] outcomes for [population]. This role is the measurement
function: it establishes whether our systems perform differently across groups,
by how much, and what to do about it.
POSITION SUMMARY
The Fairness Researcher builds and runs the quantitative evaluations that show
how our models behave across demographic and use-case slices, sets the
thresholds at which a gap becomes a blocker, and works with engineering on
mitigations that survive the next model update.
KEY RESPONSIBILITIES
•Define the fairness criteria that apply to each system and justify the choice
in writing, since the common metrics conflict with one another
•Build slice-level evaluation pipelines and run them on [every release /
a fixed cadence]
•Handle the demographic data problem honestly: what we can collect, what we can
infer, what we must not, and how to measure without either
•Investigate reported disparities end to end, from data collection through
labeling, features, objective, and threshold
•Propose and test mitigations, then measure what each one costs in overall
performance
•Maintain regression tests so a fixed disparity does not return silently
•Document methods and results so a third party could reproduce them
For regulated customers and procurement questionnaires: jurisdiction mapping, control requirements, review gates, and board briefings. If the work is closer to legislative and regulatory analysis than to product, compare it against the policy analyst templates.
AI Policy and Governance Researcher Job Description
AI POLICY AND GOVERNANCE RESEARCHER JOB DESCRIPTION
Company: __ ([City, State] / Remote)
Reports to: [General Counsel / Chief Compliance Officer / Head of Trust]
Employment type: Full-time
FLSA status: Exempt, learned professional (see classification note)
Compensation: $_ to $_ per year
ABOUT THIS ROLE
[Company Name] sells [product] into [regulated industry / multiple states / the
EU]. AI rules are arriving unevenly, and our customers ask governance questions
in every procurement cycle. This role owns the answer: what applies to us, what
we do about it, and what we can prove.
POSITION SUMMARY
The AI Policy and Governance Researcher tracks the legal and standards landscape
that reaches our products, translates it into internal requirements, builds the
governance artifacts that support procurement and audit, and advises product
teams before commitments are made.
KEY RESPONSIBILITIES
•Maintain a written map of the AI rules that apply to us by jurisdiction,
customer type, and use case, with effective dates and open questions
•Convert external frameworks into internal control requirements with named
owners and evidence
•Own our AI governance documentation set: policy, acceptable use, incident
response for model failures, and the review gate before launch
•Run the internal review board or approval process for [high-impact use cases]
•Answer AI governance sections of security questionnaires, RFPs, and vendor
assessments
•Assess third-party models and vendors we build on, including their terms,
their disclosures, and what they will not tell us
•Prepare [leadership / board] briefings on regulatory exposure and pending
changes
•Work with [legal / engineering / sales] so the same answer appears in the
contract, the product, and the documentation
REQUIRED QUALIFICATIONS
•[JD / master's degree / PhD] in law, public policy, information policy, or a
related field, or equivalent applied experience
•[Number] years in technology policy, privacy, compliance, or regulatory
analysis, with direct exposure to AI or automated decision systems
•Technical literacy sufficient to read a model card and ask the right follow-up
•Precise writing under uncertainty: able to state what is settled, what is
proposed, and what is a judgment call
•Track record of shipping a policy that teams actually followed
CLASSIFICATION AND COMPLIANCE NOTE
A policy researcher whose primary duty is advanced analysis and judgment is
generally an exempt learned professional when the salary tests are met. Note
that a governance role is not the same as legal advice: unless the person is a
licensed attorney acting in that capacity, work product is not privileged, and
internal risk memos can be discoverable. Decide deliberately what runs through
counsel and what does not, and write the routing down. This is general
information, not legal advice.
EEO STATEMENT
[Company Name] is an equal opportunity employer and provides reasonable
accommodations for the essential functions of this role.
COMPENSATION AND HOW TO APPLY
Compensation: $_ to $_ per year, [equity], [benefits summary]
To apply, email __ with your resume and a one-page memo on
one AI rule you think is most likely to reach a company our size.
Template 4: AI Safety and Red Team Researcher
For generative and agentic features: attack the system deliberately, turn findings into regression tests, and set release criteria that hold under pressure rather than under demonstration.
AI Safety and Red Team Researcher Job Description
AI SAFETY AND RED TEAM RESEARCHER JOB DESCRIPTION
Company: __ ([City, State] / Remote)
Reports to: [Head of Security / CTO / Head of Applied Research]
Employment type: Full-time
FLSA status: Exempt, learned professional (see classification note)
[customer type]. This role attacks it on purpose: finds the inputs that make it
behave badly, characterizes the failure, and stays with it until the fix holds.
POSITION SUMMARY
The AI Safety Researcher probes our systems for harmful, unsafe, and
policy-violating behavior, builds reproducible test suites from what is found,
and works with engineering on mitigations, then verifies the mitigations under
adversarial pressure rather than happy-path testing.
KEY RESPONSIBILITIES
•Run adversarial testing of [models, prompts, agents, tool use, retrieval] for
[harmful content, prompt injection, data exfiltration, jailbreaks,
overreliance, unsafe automation]
•Build a reproducible evaluation suite from discovered failures so they become
regression tests
•Define and maintain our internal content and behavior policy, including the
edge cases that decide close calls
•Assess risks specific to agentic features: what the system can do without a
human, and what should require confirmation
•Triage and investigate safety incidents reported by customers or support,
with a written postmortem
•Set release criteria for [safety-relevant features] and hold them
•Coordinate with [security] on the overlap between safety and traditional
application security
•Maintain a responsible disclosure path for external reports
REQUIRED QUALIFICATIONS
•[Bachelor's / master's / PhD] in a technical field, or equivalent demonstrated
experience in security research or applied machine learning
•[Number] years of adversarial testing, security research, or model evaluation
•Hands-on ability to build tooling and automate attacks rather than testing by
hand
•Sound judgment about what is a real risk and what is a demonstration
•Discretion with sensitive findings and disciplined disclosure practice
CLASSIFICATION AND COMPLIANCE NOTE
This role is normally exempt as a learned professional or, where the work is
primarily systems analysis and testing, under the computer employee exemption,
which has its own alternative hourly rate test. Two operational points. Testing
that touches third-party systems, accounts, or data needs written authorization
before it starts. Safety research also exposes people to distressing content:
decide in advance what support and rotation you provide, and say so in the
posting rather than after the first month. This is general information, not
legal advice.
EEO STATEMENT
[Company Name] is an equal opportunity employer and provides reasonable
accommodations for the essential functions of this role.
COMPENSATION AND HOW TO APPLY
Compensation: $_ to $_ per year, [equity], [benefits summary]
To apply, email __ with your resume and a sanitized write
up of one failure you found and how you proved it was reproducible.
Template 5: Responsible AI Lead, Split Role
For teams too small to fund a dedicated hire, with the percentage split written into the job and a named executive required to change it. The engineering half often maps to the machine learning engineer templates.
Responsible AI Lead Job Description (Split Role, Small Team)
RESPONSIBLE AI LEAD JOB DESCRIPTION (SPLIT ROLE, SMALL TEAM)
Company: __ ([City, State] / Remote)
Reports to: [Founder / CTO / Head of Product]
Employment type: Full-time, [percentage] of the role dedicated to responsible AI
FLSA status: Set by the primary duty of the whole job (see classification note)
Compensation: $_ to $_ per year
ABOUT THIS ROLE
We are a team of [number]. We cannot hire a full-time AI ethics researcher yet,
and pretending otherwise would produce a posting nobody good would answer. This
is an honest split role: [percentage] responsible AI, [percentage] [data
science / product / engineering / compliance], with the split protected rather
than absorbed the first time a deadline slips.
POSITION SUMMARY
The Responsible AI Lead owns evaluation, documentation, and the pre-launch
review for our AI features, while carrying a defined share of [the other
function]. The role exists to make sure someone is accountable for AI risk by
name, not by committee.
KEY RESPONSIBILITIES
Responsible AI ([percentage] of time):
•Run pre-launch review for AI features against a written checklist
•Maintain the evaluation suite and run it on [each release]
•Keep model and data documentation current
•Own the AI risk register and report on it [monthly / quarterly]
•Handle customer AI governance questions with [sales / support]
[Other function] ([percentage] of time):
•[List the concrete duties of the second half of the role]
REQUIRED QUALIFICATIONS
•[Degree level and field, or equivalent applied experience]
•[Number] years of hands-on work with machine learning systems in production
•Demonstrated ability to run an evaluation and act on the result
•Comfort operating without a team, a precedent, or a template
•Willingness to hold the line on the split when the roadmap gets loud
HOW WE PROTECT THE SPLIT
•The responsible AI share is written into the job description and the review
•[Named executive] approves any temporary reallocation, in writing
•Review gates are on the release checklist, not on this person's goodwill
•We revisit the split every [six months] as the AI surface grows
CLASSIFICATION AND COMPLIANCE NOTE
Classification follows the primary duty of the entire job, not the more senior
sounding half. If the larger share of the work is routine or production support,
the role may be non-exempt regardless of the ethics component, and hours must be
tracked. Do not use a split role to justify an exemption the underlying duties
do not support. Confirm your state salary threshold, which may exceed the
federal one. This is general information, not legal advice.
EEO STATEMENT
[Company Name] is an equal opportunity employer and provides reasonable
accommodations for the essential functions of this role.
COMPENSATION AND HOW TO APPLY
Compensation: $_ to $_ per year, [equity], [benefits summary]
To apply, email __ with your resume and a note on how you
would spend the first 90 days.
Template 6: AI Ethics Advisor, Fractional or Contract
For a scoped engagement before a launch, an audit, or a funding round, with deliverables, commercial terms, and a contractor classification note that treats the risk seriously.
AI Ethics Advisor Job Description (Fractional or Contract)
AI ETHICS ADVISOR JOB DESCRIPTION (FRACTIONAL OR CONTRACT)
Company: __ ([City, State] / Remote)
Engagement owner: [Founder / CTO / General Counsel]
Engagement type: [Fixed-scope project / retainer of ____ hours per month]
Worker status: Independent contractor (see classification note)
Compensation: $_ per [hour / project / month]
ABOUT THIS ENGAGEMENT
[Company Name] is [stage and size] and ships [AI feature]. We need senior
judgment on AI risk before [launch / an audit / a funding round / a large
customer], and we do not need a full-time hire to get it. This is a scoped
engagement with defined deliverables and an end date.
SCOPE OF WORK
The Advisor delivers an independent assessment of our AI systems and governance,
with written recommendations we can act on, and remains available for
[number] hours per month of follow-up during the engagement.
DELIVERABLES
•Written risk assessment of [systems in scope], including the harms we are not
currently measuring
•Review of existing evaluation methods, with gaps and recommended additions
•Draft or review of [AI policy / model cards / review gate / risk register]
•Assessment of which AI rules reach us today and which are pending
•Prioritized remediation plan with effort estimates and owners
•[Number] working sessions with [engineering / leadership]
•Final readout to [leadership / the board]
REQUIRED QUALIFICATIONS
•[Number]+ years in AI ethics, algorithmic fairness, AI policy, or model
governance, at a senior or independent level
•Direct experience with companies at our stage, not only large organizations
•Public work, publications, or references we can review
•Ability to deliver a hard assessment in writing and defend it in a room
•No conflict of interest with our competitors or vendors during the engagement
COMMERCIAL TERMS
•Rate: $________ per [hour / project], invoiced [schedule]
•Term: [start date] to [end date], renewable by written amendment
•Confidentiality and data handling: [terms]
•Intellectual property: [ownership of deliverables]
•Insurance: [professional liability requirements, if any]
CLASSIFICATION AND COMPLIANCE NOTE
Independent contractor status must be real. It rests on the economic substance
of the relationship: an advisor who sets their own hours, uses their own tools
and methods, serves other clients, and is engaged for a defined scope looks like
a contractor. Someone who works your schedule, on your equipment, under your
supervision, indefinitely, and only for you looks like an employee no matter
what the agreement says. Misclassification carries back pay, tax, and penalty
exposure at both the federal and state level, and several states apply stricter
tests than the federal one. Put the scope in writing and do not manage the
person like staff. This is general information, not legal advice.
EEO STATEMENT
[Company Name] selects contractors without regard to any protected
characteristic and provides reasonable accommodations where applicable.
HOW TO APPLY
To propose, email __ with your rate, availability, two
references, and a short outline of how you would scope the assessment.
Exempt, Non-Exempt, or Contractor
A full-time AI ethics researcher is usually an exempt learned professional, but the exemption rests on duties and not on the title. The federal regulation on learned professionals requires a primary duty of work requiring advanced knowledge in a field of science or learning, predominantly intellectual, involving consistent exercise of discretion and judgment, and customarily acquired through a prolonged course of specialized intellectual instruction.
That description fits someone designing evaluations and making judgment calls about acceptable risk. It does not fit a role built mostly from data labeling, running someone else’s scripts, and formatting reports. The federal salary threshold is $684 per week, which is $35,568 per year, after the 2024 Department of Labor rule was vacated in November 2024 and formally rescinded on May 15, 2026. Several states set higher thresholds, and our breakdown of exempt versus non-exempt classification works through the tests.
How you hire
Typical classification
What to watch
Full-time researcher
Exempt learned professional in most cases
Advanced knowledge and judgment is the test, not the job title
Split role inside an existing job
Follows the primary duty of the whole job
The ethics half does not convert an otherwise non-exempt role
Evaluation analyst or research assistant
Often non-exempt
Track hours; labeling and running scripts is not advanced knowledge work
Fractional advisor on retainer
Contractor only if genuinely independent
You set the scope, not the schedule, the tools, or the method
Academic collaborator on a grant
Usually employed by the institution
Use a research agreement and settle IP and publication rights in writing
Consultancy engagement
Their employee, not yours
Confirm confidentiality and insurance; you still own the outcome
The contractor question deserves particular care here, because fractional AI ethics advisors are common and the arrangement drifts. Someone engaged for a defined scope who sets their own hours, uses their own methods, and serves other clients looks like a contractor. Someone who works your schedule, under your supervision, indefinitely, and only for you does not, whatever the agreement says.
The Rules You Are Hiring Against
There is no single federal AI employment statute, so what governs this work is existing anti-discrimination law plus a fast-moving state, local, and international layer. That combination is the strongest argument for the hire, and it is also what the person will spend a third of their time tracking.
Two framing points matter before the detail. Existing law applies to automated decisions exactly as it applies to human ones, and the employer stays responsible even when a vendor built the tool. And the specific AI statutes are volatile: effective dates in this area have moved repeatedly, so treat any date as something to re-verify rather than to rely on.
Federal posture changed, the law did not
The Equal Employment Opportunity Commission removed its two AI technical assistance documents from its website in 2025, the May 2022 guidance on the Americans with Disabilities Act and the May 2023 guidance on assessing adverse impact in software, algorithms, and artificial intelligence. Executive Order 14281, signed April 23, 2025, further directed agencies to deprioritize disparate impact enforcement. What changed there is agency posture, not statute. Title VII of the Civil Rights Act of 1964 still prohibits selection procedures that produce unjustified adverse impact, the Americans with Disabilities Act still requires reasonable accommodation in screening, and private plaintiffs and state agencies can still bring claims. An employer remains responsible for the tools it uses, including tools built by a vendor. Hire the researcher for the law, not for the guidance page. This is general information, not legal advice.
The state patchwork is where the deadlines are
New York City has enforced Local Law 144 since July 5, 2023, requiring an independent bias audit and candidate notice for automated employment decision tools used in hiring or promotion in the city. Illinois amended its Human Rights Act through House Bill 3773, effective January 1, 2026, prohibiting AI that produces discrimination on a protected basis, banning zip code as a proxy, and requiring notice when AI is used in covered employment decisions, though the state agency withdrew its proposed implementing rules while the statutory duties stayed in force. Colorado went the other way: Senate Bill 189, signed May 14, 2026, replaced the original risk-based Colorado AI Act with a narrower disclosure regime and moved the effective date to January 1, 2027. Map the jurisdictions you actually touch, then re-check the dates, because they move.
The EU timeline shifted, but not all of it
If you sell into Europe, the AI Act still reaches you, and the schedule was recently rewritten. Regulation (EU) 2026/1744, the Digital Omnibus on AI, was published in the Official Journal on July 24, 2026 and entered into force on July 27, 2026, deferring obligations for standalone high-risk systems listed in Annex III from August 2, 2026 to December 2, 2027, and for AI embedded in products already covered by EU product safety law to August 2, 2028. Most of the Article 50 transparency obligations were not deferred and applied from August 2, 2026. Employment-related AI systems fall in the high-risk annex, so a company selling HR or hiring software into the EU has more time to prepare and no reason to stop preparing. Verify the current position before relying on any date here.
Frameworks give you the vocabulary customers use
The National Institute of Standards and Technology published the AI Risk Management Framework in January 2023 as a voluntary structure organized around four functions: govern, map, measure, and manage. It carries no force of law, which is exactly why it is useful to a small company. It gives you a shared vocabulary for procurement questionnaires, a defensible way to show a customer that your process exists, and a structure a first researcher can implement without inventing one. NIST added a Generative AI Profile in July 2024 for teams shipping generative features. Ask candidates whether they have implemented a framework at a small company rather than audited one at a large company, because the two jobs share a name and almost nothing else.
For a shared vocabulary that survives all of this churn, the NIST AI Risk Management Framework is the practical starting point, and New York City publishes plain guidance on its automated employment decision tools rule that is worth reading even outside the city, because it shows what an auditable process looks like in concrete terms. If AI touches your own hiring rather than your product, our guides to AI screening and AI in recruitment cover the employer side of the same questions.
What to Pay an AI Ethics Researcher
No Bureau of Labor Statistics occupation matches this job title, so there is no single median to quote and anyone offering one is guessing. The honest approach is to benchmark against the nearest classification for the duties you actually wrote, which is why deciding the quadrant first pays off twice.
Nearest BLS Classifications, National Medians
There is no BLS occupation for AI ethics researcher. According to the Bureau of Labor Statistics Occupational Employment and Wage Statistics survey (May 2025), median annual wages were $140,300 for computer and information research scientists, $120,230 for data scientists, $142,080 for political scientists, $106,030 for sociologists, $80,730 for compliance officers, and $80,260 for postsecondary philosophy and religion teachers. Computer and information research scientists ranged from $82,200 at the tenth percentile to $230,630 at the ninetieth (U.S. Bureau of Labor Statistics, OEWS national estimates).
Nearest BLS occupation
National median (BLS OEWS, May 2025)
When it is the right benchmark
Computer and information research scientists
$140,300 per year
Technical researcher building new evaluation methods
Political scientists
$142,080 per year
Senior policy research and regulatory analysis
Data scientists
$120,230 per year
Applied fairness and evaluation inside a product team
Sociologists
$106,030 per year
Sociotechnical impact assessment and qualitative harm research
Compliance officers
$80,730 per year
Governance, documentation, and audit readiness over research
Philosophy and religion teachers, postsecondary
$80,260 per year
Reference point for the humanities half of the candidate pool
Read the spread rather than any single number. The gap between the compliance officer median of $80,730 and the computer and information research scientist median of $140,300 is roughly $60,000, and it maps almost exactly onto the split between the governance half of this role and the technical half. Percentile range matters as much: the same survey put computer and information research scientists at $103,570 at the twenty-fifth percentile and $188,700 at the seventy-fifth, which is the band your offer has to sit in credibly.
Companies Using FirstHR Onboard 3x Faster
Join hundreds of small businesses who transformed their new hire experience.
Publish a range. This candidate pool is concentrated in metros with pay transparency requirements, moves between technology and academia where posted ranges are normal, and treats a missing range as a signal about how the company handles uncomfortable conversations. Our guide to pay transparency laws covers where a good-faith range is legally required.
Screening for Real Ethics Work
Screen for evidence that the candidate has finished something, not for the vocabulary. This field has a large gap between people who can discuss fairness fluently and people who have shipped an evaluation, argued for a threshold, and lived with the consequences of the number they chose.
The single most useful exercise is to hand over a real artifact and ask what is wrong with it: a model card, an evaluation report, or a description of one of your own features. Strong candidates ask what decision the system informs, who is affected, what data was used, and what the base rates look like, before offering any opinion. Weak candidates start with principles.
Do Not Require a Publication Record You Cannot Use
Requiring peer-reviewed publications is the most common copied requirement in this role and one of the least useful for a small company. It filters out practitioners from industry, security research, and compliance who would be strong hires, and it attracts candidates who expect research time you do not have and cannot fund. If you genuinely want the person to publish, say what for, how much time they get, and who approves it. If you do not, remove the requirement rather than leaving it in as a quality signal, because it is not one.
Reference checks are worth more here than in most technical hiring. Ask a previous manager one question: describe a time this person raised a concern that was inconvenient, and what happened next. The answer tells you whether you are hiring someone who will hold a line, and whether the previous employer let them.
Hiring an AI Ethics Researcher Without an HR Department
Small company hiring for this role fails in three predictable places: the posting is copied from a research lab, the authority is undefined, and the onboarding paperwork lands on a founder who is already the hiring manager. Each has a fix.
You copied a job description written for a research lab with three hundred people in it
Most AI ethics researcher postings online originate at large labs and consultancies. They ask for a doctorate, a publication record at peer-reviewed venues, and experience leading cross-functional programs, because that is what those organizations hire. Paste that into a posting for a company of twenty and two things happen. Candidates who could do your job assume they are underqualified and do not apply. Candidates who match the text arrive expecting a research agenda, a budget, and colleagues, and find one model, one engineer, and a launch in three weeks. Write for the work you have: which systems, which harms, which decisions this person gets to make. Specificity attracts the people who want that job and repels the people who want a different one, which is the entire purpose of a job description.
The role has responsibility and no authority, so the good hire quits in eight months
An ethics researcher who can write a report but cannot stop a launch is decoration, and experienced candidates screen for that in the first interview. Before you post, decide three things and put them in writing: which review gate this person owns, what happens procedurally when they raise a blocking concern, and who overrules them and on what record. A concrete answer, even a modest one, beats an inspiring mission statement. If the honest answer is that they advise and leadership decides, say that plainly in the posting. You will get fewer applicants and better fits, and you will not lose a senior hire six months in over a promise nobody remembers making.
The compliance paperwork behind an AI governance hire lands on the founder
This hire generates more onboarding than a typical one: confidentiality and invention agreements, data access approvals tied to sensitive datasets, security and privacy training before the first query, a signed acknowledgment of the internal AI and acceptable use policy, disclosure and publication terms if the person expects to write publicly, and named review gates that need to exist in a system rather than in somebody’s memory. FirstHR was built for that kind of sequence at companies without an HR department. The onboarding wizard runs the same steps for every hire, e-signature handles the agreements and policy acknowledgments, document management stores the signed records against the employee profile with renewal dates attached, and training modules deliver security and policy orientation before day one. Applicant tracking is coming soon to FirstHR.
Once the offer is signed, the work becomes a repeatable sequence rather than a scramble, and our onboarding checklist covers the general version. For this hire specifically, the sensitive-data access approvals and the policy acknowledgments are the two steps most often completed late, and both are the ones an auditor asks to see first. More AI hiring templates sit in the hiring templates library.
Key Takeaways
AI ethics researcher covers four different jobs: measurement and evaluation, policy and governance, safety and red teaming, and academic research. Decide which one you are buying before writing the posting, because the candidate pools barely overlap.
No BLS occupation matches the title, so benchmark against the nearest classification for the duties you wrote: nearest May 2025 medians run from $80,730 for compliance officers to $142,080 for political scientists.
A full-time researcher is usually an exempt learned professional, but a role built from labeling and running scripts is non-exempt, and a split role follows the primary duty of the whole job.
Fractional advisory engagements are common and drift toward employment; keep the scope defined and stop short of directing the schedule, tools, and method.
Federal AI hiring guidance was withdrawn in 2025, but Title VII and the ADA still apply to automated selection procedures, and the employer stays responsible for a vendor’s tool.
State and international deadlines move: New York City has enforced its bias audit rule since July 5, 2023, Illinois since January 1, 2026, Colorado now from January 1, 2027, and the EU deferred high-risk obligations to December 2, 2027.
Define the review gate and the escalation path in the posting; an ethics hire with responsibility and no authority leaves within a year.
An AI governance hire generates more paperwork than most: confidentiality and invention agreements, data access approvals, security training before the first query, and a signed acknowledgment of your internal AI policy. FirstHR runs that sequence the same way every time, with e-signature for the agreements, document storage against each employee profile, and renewal dates tracked so nothing lapses quietly. Applicant tracking is coming soon to FirstHR.
Frequently Asked Questions
What does an AI ethics researcher do?
An AI ethics researcher identifies, measures, and documents the harms an AI system can cause, then turns those findings into changes the engineering team ships. The work splits into four fairly different activities: quantitative evaluation of model behavior across groups and edge cases, policy and governance work that maps external rules to internal requirements, adversarial safety testing of generative and agentic features, and academic research that produces new methods. Most small companies need the first three and almost never need the fourth. Day to day the job looks like designing an evaluation, defining what result counts as unacceptable, running it, investigating a disparity through the data and the objective function, writing a model card, and answering the AI governance section of a customer questionnaire. Decide which of those four activities you are actually buying before you write the posting, because the candidate pools barely overlap.
What qualifications does an AI ethics researcher need?
It depends on which half of the role you are hiring, and the honest answer for most small companies is less than the copied postings claim. Evaluation and fairness work needs applied statistics, working fluency in Python and SQL, and enough machine learning background to trace a disparity from data collection through labeling, features, objective, and threshold. Policy and governance work needs regulatory analysis, precise writing under uncertainty, and enough technical literacy to read a model card and ask the right follow-up question. Safety and red teaming often draws its best candidates from security research rather than from ethics. A doctorate is standard at large research labs and frequently unnecessary elsewhere. Ask for a degree only when the work genuinely requires it, and mark it required or preferred explicitly so candidates can self-select instead of guessing.
How much does an AI ethics researcher make?
No Bureau of Labor Statistics occupation matches the title, so the figure has to come from nearby classifications. According to the Bureau of Labor Statistics Occupational Employment and Wage Statistics survey (May 2025), the national median annual wage was $140,300 for computer and information research scientists, $120,230 for data scientists, $142,080 for political scientists, $106,030 for sociologists, $80,730 for compliance officers, and $80,260 for postsecondary philosophy and religion teachers. The spread is the point: a publication-track technical researcher and a governance-focused hire are paid from different parts of the market. Computer and information research scientists ranged from $82,200 at the tenth percentile to $230,630 at the ninetieth in the same survey. Pick the nearest classification to the duties you wrote, then adjust for your metro and stage, and publish a range because this candidate pool expects one.
Is an AI ethics researcher exempt from overtime?
Usually yes, as a learned professional, but the exemption rests on duties rather than on the title. The federal regulation on learned professionals requires that the primary duty be work requiring advanced knowledge in a field of science or learning, predominantly intellectual, involving consistent exercise of discretion and judgment, and customarily acquired through a prolonged course of specialized intellectual instruction. A researcher designing evaluations and making judgment calls about acceptable risk fits that description. A role built mostly from data labeling, running someone else’s test scripts, and formatting reports does not, and is non-exempt and owed overtime past 40 hours in a week. The federal salary threshold is $684 per week, which is $35,568 per year, after the 2024 rule was vacated in November 2024 and formally rescinded on May 15, 2026. Several states set higher thresholds, so check yours. This is general information, not legal advice.
Does a small company need a full-time AI ethics researcher?
Often not, and the split role or fractional templates on this page exist for exactly that reason. The realistic trigger for a dedicated hire is a combination of factors: your AI makes or materially influences decisions about people, you sell into regulated industries or jurisdictions with automated decision rules, customers now ask governance questions in procurement, and no single person can currently answer them. Below that threshold, two structures work better than a stretched full-time posting. The first is an honest split role with the responsible AI percentage written into the job description and a named executive required to approve any reallocation. The second is a scoped fractional engagement before a launch, an audit, or a funding round, with defined deliverables and an end date. Both are better than a full-time posting you cannot actually fund or fill.
What is the difference between an AI ethics researcher and an AI policy analyst?
The researcher measures what the system does; the policy analyst determines what the rules require. A fairness or safety researcher works inside your data and models: building evaluation pipelines, choosing metrics, running adversarial tests, and investigating why a subgroup gap exists. The output is empirical. A policy or governance analyst works outside the model: mapping which laws and standards reach your product by jurisdiction and use case, converting them into internal control requirements, and producing the documentation that procurement, auditors, and boards ask for. The output is a written position. Larger organizations staff both. Small companies hire one person who leans one direction, which is why guessing wrong is expensive. Write down which output you need first, then use the matching template rather than a posting that asks for both and attracts neither.
What laws govern AI use in hiring and employment?
There is no single federal AI employment statute, so the answer is a stack of existing law plus a growing state and local layer. Title VII of the Civil Rights Act of 1964, the Americans with Disabilities Act, and the Age Discrimination in Employment Act all apply to automated selection procedures exactly as they apply to human ones, and the employer stays responsible even when a vendor built the tool. On top of that sit specific rules: New York City has required independent bias audits and candidate notice for automated employment decision tools since July 5, 2023, and Illinois amended its Human Rights Act effective January 1, 2026 to prohibit AI-driven discrimination, ban zip code proxies, and require notice. Colorado moved its own law to January 1, 2027 through Senate Bill 189. Map your jurisdictions and re-verify the dates, because they change. FirstHR is an onboarding and HR platform, not a payroll provider. Applicant tracking is coming soon to FirstHR.
