EmergingSynthetic data and environment generation

Privacy-Preserving Synthetic Data Scientist

Also called Private Synthetic Data Researcher

A privacy-preserving synthetic data scientist tests whether artificial datasets leak information about real people, and whether they still work for their purpose. That lets a company share and train on data it could not otherwise use.

What they do

The work starts with a synthetic dataset and a question: what could an attacker learn from this? They design and run membership inference, attribute inference and re-identification tests against the generator and its outputs. They set up the experiments, choose baselines and record every attack assumption. Alongside that, they measure utility on the downstream tasks the data is meant for. They tune the generator, add noise or constraints, and rerun the tests to see the trade-off. They check for memorised records and rare-value leakage. The results go into a report that engineers, legal and product can all read.

What a strong hire delivers

In the first months, a strong hire sets up a repeatable attack test suite that runs against every new synthetic release. They agree with the business what residual risk is acceptable and document it. They produce a first evaluation and utility report on a real dataset, with clear recommendations. They also leave behind a way for the generation team to see the privacy and utility trade-off before data ships, not after.

How you can tell they are good

Ask them to walk through a past evaluation end to end. A good candidate names the threat model first, explains which attacks they ran and why, and says what the tests could not rule out. They talk about utility in terms of tasks, not generic similarity scores. They are honest that low measured risk is not zero risk. Be wary of anyone who treats a single privacy metric as proof of safety.

Why it is hard to fill

Almost nobody carries this title. The work sits inside machine learning research, privacy engineering, statistics and data science teams, often as a side duty. Candidates tend to be strong on one side only: generative modelling without attack experience, or privacy knowledge without hands-on generator work. The mix is rare, and people who have it often do not describe themselves this way. Titles and CVs rarely signal it, so searching takes judgement.

Where to find them

Look at privacy engineering and security research teams, applied research groups working on generative models, and statistics or biostatistics teams that handle sensitive records. Test engineers and simulation specialists can move across well. Healthcare, finance and public sector data teams have done this work for years under other names. Frontier labs and model providers also have people doing it inside safety or data teams.

Indicative pay

Not enough published evidence yet to give a fair range for this role, which is common for work this new. We benchmark it for you as part of a search.

Interview questions for a Privacy-Preserving Synthetic Data Scientist

  1. Describe a time you tested a synthetic dataset for privacy leakage. What threat model did you use and what did you find?

    A strong answer names the attacker's assumptions, the attacks run and the outcome, including limits. Look for specifics and honesty about uncertainty. Worry if they only cite one aggregate metric or cannot say what the attacker knew.

  2. How do you decide whether a synthetic dataset is useful enough?

    They should tie utility to the downstream task, compare against real-data baselines and check subgroups and rare cases. Worry if they rely only on distribution similarity or have no view of what the data is for.

4 more questions, with what a strong answer covers, are in the Role Kit.

Job description you can post

We build and use synthetic data so our teams can work with sensitive information safely. We need someone to test that data hard. You will attack our synthetic datasets, measure how useful they remain, and tell us plainly how much risk is left before anything is shared or used.

What you will do

  • Design and run membership inference, attribute inference and re-identification tests on synthetic data
  • Measure utility on the downstream tasks the data supports
  • Work out privacy and utility trade-offs with the generation team
  • Document threat models, attack assumptions and residual risk
  • Build repeatable evaluation pipelines that run on each new release
  • Write clear reports for engineering, legal and product audiences

What you will bring

  • Demonstrated experience evaluating generative models or synthetic data
  • Hands-on work running privacy attacks or similar adversarial tests
  • Strong experimental design and statistical judgement
  • Ability to write and defend a threat model
  • Comfortable coding evaluation tooling and working with dataset validators
  • Clear written communication of technical risk to non-specialists

Nice to have

  • Experience with differential privacy or other formal privacy methods
  • Background in simulation or digital twins
  • Work with regulated or sensitive data

Free to use and adapt. If it saves you time, a link back to this page is appreciated.

Skills that matter

  • Synthetic dataAdvancedGenerate test or training examples with validity controls
  • Experimental designAdvancedChoose comparisons that identify credible effects
  • Privacy engineeringWorkingLimit disclosure and unauthorised personal-data use
  • Simulation and digital twinsWorkingBuild useful virtual representations of environments

Other roles in synthetic data and environment generation

Hiring nowSynthetic Data EngineerA Synthetic Data Engineer builds pipelines that generate artificial training and evaluation data, then proves it is useful. For a company, that…EmergingSynthetic Data Validation ScientistA Synthetic Data Validation Scientist tests whether artificial data behaves like the real world it is meant to stand in for. That protects model…Hiring nowSimulation Scenario DesignerA Simulation Scenario Designer builds the realistic test scenarios and edge cases that show where an AI system breaks. They matter because a team…Hiring nowRL Environment EngineerAn RL Environment Engineer builds the simulated worlds and tasks in which models learn through reinforcement learning. The quality of those…

Related roles in other families

Hiring nowClinical AI ScientistClinical AI and healthcare deliveryA Clinical AI Scientist tests whether an AI tool works for a specific clinical question and patient group. They give a company the evidence to…Hiring nowClinical AI Implementation SpecialistClinical AI and healthcare deliveryA Clinical AI Implementation Specialist takes validated AI tools and puts them safely into real care workflows. They matter because a model that…EmergingClinical AI Safety OfficerClinical AI and healthcare deliveryA Clinical AI Safety Officer coordinates oversight of risks and incidents when AI is used in care. They help the company spot patient-safety…EmergingAI Medical Documentation SpecialistClinical AI and healthcare deliveryAn AI Medical Documentation Specialist configures and tests AI-assisted clinical note tools so that clinicians can trust them. They find the errors…

How we wrote this

Role definitions are our own editorial work, checked against occupational classifications, standards, research and employer pages. Titles vary by company, and this is a description of the work, not a job advert.

← Synthetic data and environment generationAll new job families

Start a search

Hiring a Privacy-Preserving Synthetic Data Scientist?

Tell us the work. We find the people already doing it, whatever their job title says, and you meet them already interviewed for your role.