AI governance is the internal system that decides how AI gets built and shipped — who approves a model, what testing it has to survive, who is accountable when it misbehaves. AI compliance is the narrower job of proving that system satisfies an external rule: a state statute, a customer contract, an auditor's request.

Governance is the machine. Compliance is the receipt. Companies routinely hire for the second when the problem is the first, which is how you end up with a compliance manager who has no authority to stop a launch.

The internal system of policies, approval gates and accountability that decides how AI gets built, bought and shipped inside a company — who signs off on a model, what testing it has to pass, and who owns the outcome if it fails. On our ladder it runs from AI Governance Analyst up through Director and VP to Chief AI Officer.

It sits above compliance and risk rather than beside them: those functions typically report into or feed a governance program, not the other way around.

It's the umbrella term for independent testing, attestation and ongoing verification of an AI system — proving it does what it should and does not do what it shouldn't, on a continuing basis rather than a one-time sign-off. It sits across governance, risk, compliance and security rather than belonging to any one of them alone.

Almost nobody in the US hires under that title yet. The work gets done by auditors, compliance managers, risk analysts and security engineers, each holding one slice of it — which is the gap this practice is built to translate.

The practice of identifying, scoring and mitigating what could go wrong with an AI system before it does — model drift, bias, security exposure, third-party model risk — using a structured framework rather than ad hoc review. There's deep overlap with banking model risk management, which is why a lot of strong candidates come from that world.

Titles run AI Risk Analyst through AI Risk Manager to Director of AI Risk or Head of Model Risk, with the mandate sometimes sitting inside an existing Chief Risk Officer's remit rather than as a standalone seat.

A discipline imported largely from banking: independently validating a model's performance, assumptions and failure modes before and after deployment, distinct from whoever built it. Inside AI governance work it now covers LLMs and agentic systems, not just the credit and fraud models it originated with.

It sits inside the Risk domain on our map, and candidates with real MRM backgrounds — especially from regulated banks — transfer into AI risk roles faster than almost any other background.

The set of controls, documentation and monitoring a company runs to prove its AI use satisfies external requirements — a state statute, a customer's security questionnaire, an auditor's request. It's the receipt, not the decision-making machine that is governance itself.

The role scales from AI Compliance Analyst up to Chief Compliance Officer depending on level. Companies that band this work as a subset of privacy or legal, rather than its own track, tend to lose the candidates who can actually run one.

Deliberately trying to break an AI system the way an attacker or bad actor would — prompt injection, jailbreaks, data exfiltration, harmful output — before it ships, and often on a recurring basis after. It's adversarial testing, not code review.

This prices against security engineering, not compliance — the ladder runs from ML Security Engineer and AI Red Teamer up through LLM Security Architect. Companies that source it as a compliance hire consistently strike out — the OWASP GenAI Security Project's LLM Top 10 is the closest thing this field has to a shared standard, and it's a security document, not a policy one.

The executive who owns AI governance end to end — policy, approval gates, risk tolerance, and increasingly the regulatory relationship — reporting to the CEO, COO or board rather than functioning as a technical role. It's the top rung of our Governance ladder, alongside Chief AI Governance Officer.

It's still a rare title: most companies route the mandate through an existing CRO, CISO or General Counsel instead of creating a dedicated seat, which is worth knowing before you assume the market has standardized on it.

Less writing policy than running a gate: reviewing new AI use cases before they launch, deciding what needs sign-off versus what's low-risk enough to wave through, coordinating with legal, security and the model owners, and reporting up on where the program has gaps. A good week has more meetings adjudicating specific launches than time spent on the governance framework itself.

The job is judgment-heavy and cross-functional by design — which is why the strongest candidates tend to have sat inside risk, compliance, audit or security first, not straight out of a policy or ethics background alone.

Fair question, and we pressure-tested it before committing. Posting volume has held flat at roughly 71 net-new US roles a week through the first seven months of 2026 — no seasonal collapse, which is the pattern hype markets show first.

There is also a direct precedent. GDPR created the Data Protection Officer, which turned into a durable specialist recruiting niche that still exists eight years later. Regulatory-created roles do not tend to evaporate; they get absorbed into the org chart.

Yes, for one specific reason: its extraterritorial reach. Any US company selling into or operating in the EU needs staff who understand its high-risk-system obligations, regardless of where that company is headquartered — a real driver of US-based compliance hiring, not a European issue happening somewhere else.

It is not, however, the primary driver in the US market. That is the state patchwork — Texas, California, Colorado — layered under enforcement from the FTC, SEC, DOJ, EEOC and FCC. A company with no EU exposure at all still faces real hiring pressure for this work.

More than a handful, and the list keeps moving. Texas (TRAIGA) and California (SB 53, AB 2013) have had laws in force since January 2026; Colorado rewrote its original framework into SB 26-189, effective January 2027; New York City's Local Law 144 has required AI hiring-tool bias audits since 2023, with enforcement tightening after a critical 2026 comptroller audit found the city's own oversight weak.

There is no single federal AI law playing the EU AI Act's role here — which is the real reason this hiring wave exists. Someone has to track a moving state-by-state patchwork instead of one static rulebook.

In a rewritten form. The original law (SB 24-205) was repealed and replaced by SB 26-189, signed in May 2026, which delayed the effective date to January 1, 2027 and narrowed the requirements considerably — dropping the mandatory risk-management-program and annual impact-assessment obligations in favor of a disclosure-and-consumer-rights framework for automated decision-making in employment and other consequential decisions.

Colorado's Attorney General has also said he won't enforce it, or its predecessor, until rulemaking is finished — so the practical compliance clock is later than the headline date suggests, but companies hiring for this now are hiring ahead of it, not behind it.

Texas's Responsible AI Governance Act, in force since January 2026, targets specific harmful uses rather than regulating AI broadly: government use of AI for social scoring or manipulation, and AI systems intentionally designed to incite self-harm or unlawful acts, among other named prohibitions. It's narrower in scope than the EU AI Act's risk-tiered model.

For hiring purposes, TRAIGA matters less for its specific bans and more as one more data point that state-level AI law is arriving unevenly — which is exactly why compliance and governance hires increasingly need to track a patchwork, not a single federal standard.

If you use an automated employment decision tool to screen or score candidates for a role based in New York City, yes: the law has required an independent annual bias audit, public posting of the results, and candidate notice since 2023. It was an early mover and is still one of the most concrete AI-hiring compliance obligations in the country.

Enforcement had been inconsistent — a 2026 city comptroller audit found the Department of Consumer and Worker Protection's oversight weak on complaint handling and review quality — but DCWP has committed to tightening it, which typically means more investigations and penalty risk ahead, not less.

There isn't a single number that works across all eight domains — or even within one of them. Company stage, location, scope, and seniority move the number more than the title on the job spec does.

An early-stage startup building its first AI governance function might offer meaningfully more equity against a lower base than a well-funded company paying top-of-market cash with thinner upside. A fully remote or secondary-market company prices differently than one anchored to a San Francisco or New York City band. Someone hired to build a program from scratch commands a different number than someone inheriting an already-mature one, even at the same title. Stack those variables together and a flat range stops being useful.

Rather than publish a band that would be wrong for most of the roles we actually work, we'd rather scope it against your specific situation — stage, location, and what the role needs to do in year one. Tell us the specifics and we'll help you land on a number that's defensible for your search, not a generic figure pulled off a website.

It depends entirely on why the role was created. Compliance-triggered hires normally sit under Legal, Risk, or the Chief Privacy Officer. Security-triggered hires sit under the CISO.

At AI vendors it is often neither — if the role exists to clear customer security reviews and unblock revenue, it belongs in go-to-market, and treating it as a compliance seat will cost you the candidates who are good at it.

Start by being honest about whether you actually need one. Most companies at this stage are better served routing the mandate through an existing CRO, CISO or General Counsel with a formal AI remit added, rather than creating a net-new C-level seat — the title is still rare enough that a search built around it alone will take longer and cost more than the problem requires.

When a dedicated CAIO genuinely is warranted — real regulatory exposure, board-level AI risk, a company built around AI as the product — the search functions like any executive search: reference-checked track record running governance at scale, not just AI fluency.

Start with an inventory, not a policy: find every AI use case already live in the company, including shadow use nobody signed off on, before writing a single rule. Most first governance programs fail because they publish a framework for AI use the company hasn't actually mapped yet.

From there: a tiered risk classification for new use cases, a lightweight intake and approval process people will actually use, and one accountable owner — even before you have headcount to build a full team, someone needs to hold the gate.

Score each AI use case against a small number of concrete dimensions — data sensitivity, decision impact on people (hiring, credit, healthcare especially), model opacity, and vendor dependency — rather than a generic checklist borrowed from traditional IT risk. AI-specific failure modes like hallucination and drift need their own criteria.

NIST's AI Risk Management Framework and ISO/IEC 23894 are the two most widely cited starting structures; most companies adapt rather than adopt either wholesale.

A standing process for anyone in the company to declare a new AI use before building or buying it — what tool, what data, what decision it touches — so governance can classify the risk tier and route it to the right level of review, instead of finding out about it after launch.

The single biggest failure mode here is making intake too heavy: if submitting a use case takes longer than just building it quietly, people build it quietly. The best programs make low-risk approval nearly instant and reserve real scrutiny for the cases that actually warrant it.

Documentation first: model inventory, intended use versus actual use, testing and validation records, and a clear map of who approved what and when. Auditors — internal or independent — are checking whether your stated controls match what's actually happening, and the gap between the two is where most findings come from.

If this is your first AI audit, run an internal dry run against your own policy before an external one arrives; it's far cheaper to find your own gaps than to have an auditor find them for you.

What data does the tool train on or retain, and does that include your company's data. What happens when the model is wrong — is there a human review step before a consequential decision is made. Has the vendor had an independent bias or security audit, and will they share it. What's their incident-notification commitment if something goes wrong post-deployment.

If the vendor can't answer these clearly, that's itself the answer — and it's exactly the kind of gap an AI compliance or vendor-risk hire exists to catch before procurement signs anything.

Three main lanes: existing privacy, compliance or audit professionals who picked up AI-specific scope as their company's exposure grew; risk and model-risk people from regulated industries, especially banking, where validating a model before deployment is already a mature discipline; and security engineers who moved into AI-specific red teaming or LLM security as that specialty emerged.

Very few came into the field with an "AI governance" title from day one — almost everyone doing this work well translated in from an adjacent discipline, which is exactly why résumé titles alone are a poor filter for this market.

For someone already in privacy, compliance, risk, audit or security, yes — it's a specialization that adds scope and compensation to a track record you already have, in a function that's hiring steadily rather than cyclically: postings held around 71 net-new US roles a week through the first seven months of 2026, without the seasonal collapse hype-driven markets show first.

As a first job with no adjacent background, it's harder — there isn't yet an established junior-to-senior pipeline built for people entering cold, so most strong candidates arrive via one of the adjacent disciplines rather than a dedicated AI governance track.

The fastest path is adding AI-specific scope to work you're already doing — leading the review of one AI use case, sitting in on a vendor security assessment for an AI tool, or volunteering to help draft an AI acceptable-use policy — rather than waiting for a title change first. Hiring managers consistently look for demonstrated exposure to AI-specific risk, not a certification alone.

IAPP's AIGP and similar credentials help signal intent, but they're rarely sufficient on their own; pair one with a real example of AI-specific work you've actually done.

A generalist agency spreads across dozens of functions and rotates your search through whichever recruiter has bandwidth that week — often someone who can't tell an AI risk analyst from an AI compliance analyst, let alone which one your role actually needs. This practice works one market: eight domains, the real title ladder inside each, and the comp bands that go with them, published openly rather than guessed at.

Every search here is run directly, without handing you off to an account manager once the contract is signed. That's less a promise than a function of staying this specialized — there isn't a bench of generalist recruiters to hand it to.

It starts with a real conversation about the problem you're trying to solve, not a job-description intake call. From there we run a targeted search against our own market map, bring you a short list of vetted candidates with context on why each one fits, and stay involved through offer and close rather than disappearing once a resume lands in your inbox.

There's no hand-off and no volume play — a short, curated list beats a hundred inbound resumes, and we're still on the call when an offer gets negotiated, not just when the introduction happens.