One Job, Five Titles

AlphaSense has a role open called Compliance Analyst, AI Governance. The first sentence of the description reads: "The AI Governance Analyst is central to building and operating AlphaSense's AI governance program."

The employer could not hold one name for the job between the title bar and the opening line.

I do not point that out to embarrass anyone. I point it out because it is the clearest small example of something that costs candidates and employers real money, and almost nobody names it. A job board in a young field has a description problem before it has a distribution problem.

The same work, filed five ways

Take one duty: keep an inventory of the AI a company is using, assess each new use case against a template, and be able to produce the evidence when a regulator, an auditor or a customer asks for it.

That duty appeared in 2026 under all of these titles.

The TransUnion and Carrington postings are close to twins. TransUnion asks the analyst to "manage the AI inventory" and to "maintain a global AI assessment template." Carrington asks for recommendations on governance documentation "including policies, standards, procedures, model inventories, use-case registers, and control mappings," and for support with "evidence collection" for "internal audits, external audits, due diligence, and regulatory examinations." The advertised midpoints sit about five thousand dollars apart. The titles share not one word.

Two of those five never say "AI governance" anywhere in the title. A candidate typing that phrase into a search box will not see them.

Why it happens

The easy explanation is laziness, and I do not think that is it. The real reason is structural, and the IAPP has measured it.

In its AI Governance Profession Report, published in April 2025 and drawn from more than 670 respondents across 45 countries and territories, the IAPP asked which function owns AI governance inside the organisation. Privacy said 22 percent. Legal and compliance said 22 percent. IT said 17 percent. Data governance said 10 percent. Ethics and compliance said 6 percent. Security said 5 percent.

Six different departments own the same work. Each names it in its own dialect. A privacy team calls it governance, a bank calls it model risk, a defence contractor calls it responsible AI, a hospital files it under compliance. Every one of those choices is coherent from inside the building. The incoherence is only visible from outside, where somebody is typing one phrase and seeing a fraction of what they could do.

The frameworks are not old enough to have settled the language either. The NIST AI Risk Management Framework was released in January 2023 and is voluntary, so everyone who adopts it interprets it a little differently. ISO/IEC 42001 was published in December 2023. The EU AI Act entered into force in August 2024 and only became generally applicable last month. When the standards are two to four years old and still moving, the job titles move with them.

It is not just practitioners who struggle with this. When Indeed's Hiring Lab tried to measure the growth of this field in 2025, it could not use one label. It had to bundle five: responsible AI, ethical AI, AI ethics, AI governance, and AI safety. A research team with good data hit the same wall a job seeker hits.

The academic work says the same thing more carefully. Shalaleh Rismani and AJung Moon, in "What does it mean to be a responsible AI practitioner: An ontology of roles and skills", built an ontology of responsible AI roles from job postings and practitioner interviews. Their conclusion: "due to the nascent nature of these roles, however, it is unclear to future employers and aspiring AI ethicists what specific function these roles serve and what skills are necessary to serve the functions."

The honest objection

Someone will say that every profession has messy titles and this is unremarkable. That deserves an answer rather than a dismissal.

Two things are true. Title churn is happening across all of AI, not only in governance. Indeed found that the number of AI-touched job titles in the US went from 264 in early 2022 to 822 in early 2026. Some of that is simply a young industry naming things.

And some of the variation in AI governance is real rather than cosmetic. Within the same bucket you will find an analyst advertised at $79,000 and a governance attorney advertised above $300,000. Those are genuinely different jobs, and the titles are doing useful work in telling them apart. Anyone who claims all the variance is noise is overselling.

So the narrow claim is the one worth making: at the analyst tier, substantially the same duties are being advertised under names with no common vocabulary, and candidates are missing roles because of it.

What actually fixes it

Three things, and none of them are exciting.

The first is a controlled vocabulary. Agree a preferred term, map the variants to it, and let the system do the translation. If somebody searches "algorithmic audit" and the preferred term is "AI assurance," they should get the right results without ever learning that a substitution happened. This is old library technology. It works, and it has worked since long before anyone was governing a model.

The second is describing roles by what the person does rather than by which department signs the offer letter. Someone who maintains model documentation and produces evidence for auditors is doing recognisably the same job whether they report to the Chief Risk Officer or the General Counsel. Reorganisations move departments around. They do not change the work.

The third is being honest about the gaps. When you classify a body of material properly, you end up with empty categories, and there are two very different reasons a category can be empty. Either nobody has looked, or people have looked and found nothing. Those are not the same finding and they should not look the same. A job board inherits that obligation. If the frameworks say a role should exist and nobody is hiring for it, that is worth saying plainly rather than returning an empty result and letting the candidate assume they searched wrong.

The question worth asking

So which word do you type? And if you guess wrong, does anyone ever tell you what you missed?

The useful measure for a job board in this field is not how many people saw a listing. It is whether the person who could have done the job found it using the words they actually had.

That is a description problem. It comes first, and no amount of distribution fixes it.

Sources. Postings: AlphaSense via Greenhouse (job 8648659002); TransUnion and Carrington Mortgage Services via Built In; New York Power Authority careers site; USAJOBS announcement 873842400 for the Office of Management and Budget. Salary figures exactly as advertised. Research: IAPP AI Governance Profession Report, April 2025, 670+ respondents across 45 countries and territories; Indeed Hiring Lab, "The Rise of Responsible AI Jobs" (Gallacher and Aoki, June 2025) and "AI is no longer just a tech occupation story" (Adrjan, July 2026); Rismani and Moon, arXiv:2205.03946. Frameworks: NIST AI RMF 1.0 released 26 January 2023; ISO/IEC 42001 published 18 December 2023; EU AI Act entered into force 1 August 2024 and became generally applicable 2 August 2026. Postings cited were advertised during 2026; several have since closed.