Hundreds of applicants pile into generic "AI safety" postings while adversarial red-teaming and integrity engineering roles sit underfilled — and pay up to $1 million.

OpenAI's Integrity team, housed inside Applied Engineering, employs Machine Learning Engineers in San Francisco at $266,000–$555,000 plus equity. Their job is to identify misuse, prevent abuse, and protect users (a role that barely existed as a named discipline three years ago). Zero G Talent's board lists 53 OpenAI roles added in the past week alone, including an Agentic Risk Analyst slot paying $288,000–$425,000.

That single job band captures a broader transformation. AI safety has splintered from a vague, monolithic aspiration into at least three distinct career tracks, each with its own skill stack, employer ecosystem, and compensation logic. The professionals who understand this map are accessing roles paying $160,000–$340,000 without fighting through the most crowded applicant pools. Everyone else is still applying to a job that no longer exists.

A Workforce Explosion and a Talent Mismatch

The global AI safety workforce grew from roughly 400 full-time equivalents in 2022 to over 1,100 in 2025, split between about 600 technical safety roles and 500 non-technical positions. Hiring for AI trust and safety roles grew 36% year-on-year, and Quess IT Staffing projects a further 15–25% increase ahead. Overall market growth of 30–40% is expected over the next one to two years.

Yet roughly 79% of employers report difficulty finding qualified candidates. In some markets, only one qualified engineer is available for every 10 generative AI roles. This is not a cyclical shortage that will correct itself. The field has differentiated faster than the labor supply can adapt. Three distinct hiring lanes have formed. Most job-seekers still see only one.

The result is a paradox: hundreds of applicants cluster around generic "AI safety" postings while specialized sub-domains (adversarial red-teaming, production integrity engineering, regulatory governance) go underfilled. The professionals who figure out which lane matches their existing skills can bypass the bottleneck entirely.

The Adversarial Red-Teamer

Red-teamers probe AI systems for failure modes. They discover jailbreaks, emergent deception, and misuse pathways before deployment. It is the most technically elite archetype, drawing on security research, alignment theory, and a kind of creative adversarial thinking that resists easy credentialing.

The work is concentrated in a handful of elite labs. Anthropic, OpenAI, Google DeepMind, and Thinking Machines Lab (the startup led by former OpenAI CTO Mira Murati, where technical staff earn $450,000–$500,000) all employ dedicated red-teamers. Safe Superintelligence Inc., co-founded by former OpenAI chief scientist Ilya Sutskever, pays technical staff $150,000–$500,000. Senior technical safety researchers at top labs command $500,000–$1,000,000+ when equity is included.

The skill demands are steep. Fluency in ML internals, prompt injection techniques, and often formal methods is table stakes. Training pipelines like MATS (a 12-week mentorship program with a 4–7% admission rate) and ARENA, a 4–5 week intensive with an open curriculum, serve as key on-ramps. Zero G Talent's board shows 41 Anthropic roles added in the past week, including a Product Manager, Safeguards Rare Harms position paying $305,000–$385,000.

Geographic hubs are narrow: the San Francisco Bay Area, London (home to DeepMind, GovAI, and the UK AI Safety Institute), and Washington D.C. The concentration reflects both the clustering of frontier labs and the classified-adjacent nature of the work. This is not a remote-friendly archetype.

The Integrity Engineer

If red-teamers find the cracks, integrity engineers build the guardrails. They construct the classifiers, monitoring systems, and intervention layers that translate safety research into production infrastructure. This archetype accounts for the largest volume of new hiring.

Anthropic's Trust and Safety roles carry a median total compensation of $198,588. AI Security Engineers in the USA earn $185,000–$248,000 at mid-level and $298,000+ at Lead and Staff levels, a premium on engineers who can ship production-grade safety tooling rather than theorize about it.

The skill stack is distinct from red-teaming. It requires MLOps fluency, CI/CD pipeline integration, real-time monitoring architecture, and the ability to work alongside product teams. These are not researchers in a lab. They are engineers embedding safety into systems that serve millions of users.

Employers span frontier labs, enterprise SaaS companies, and increasingly BFSI firms. Banking, financial services, and insurance companies are expected to see a 25–30% rise in demand for these positions as regulated industries deploy generative AI at scale. The hiring base is widening, and the roles are multiplying.

Governance: From Advisory to Mission-Critical

Governance professionals have moved from advisory to mission-critical. The driver is regulatory deadlines. The EU AI Act threatens fines of up to €35 million or 7% of global turnover for noncompliance. NIST released its AI Risk Management Framework (AI RMF 1.0) on January 26, 2023, and published a Generative AI Profile on July 26, 2024. On April 7, 2026, NIST released a concept note for an AI RMF Profile on Trustworthy AI in Critical Infrastructure. Job postings referencing "NIST AI RMF" number in the hundreds on LinkedIn, with 822 listings on Indeed and 957 on Glassdoor, broad private-sector adoption of a framework that barely existed three years ago.

Governance roles are now expected to integrate into MLOps workflows and CI/CD pipelines rather than operate as external reviewers. The line between compliance and engineering is blurring. Key titles include Responsible AI Lead, AI Governance Manager, AI Compliance Officer, and AI Risk Manager. Salary increases of 30–40% are being observed for skills in compliance frameworks and AI risk expertise. CISO31K alumni (professionals with ISO 31000 risk manager certification) report a median 47% salary lift across Risk Manager, ERM, and CRO tracks.

Major employers include the Big Four and consultancies. Infosys has created a dedicated Responsible AI office. KPMG India is expanding its responsible AI team to cover governance, risk management, ethical AI policy, and AI security and privacy. Accenture maintains a Chief Responsible AI Officer. Federal agencies are also hiring: the US government technical salary ceiling has been adjusted to $197,200 to compete for expert talent, and NIST launched its Trustworthy and Responsible AI Resource Center on March 30, 2023.

This archetype is the most accessible to non-technical career switchers. Domain expertise in regulated industries (BFSI, healthcare) combined with NIST AI RMF fluency and a policy fellowship like the Horizon Fellowship (which places candidates in US federal policy roles) or TechCongress (which places technologists in Congressional offices) can open the door.

Compensation: Why Mapping Your Archetype Matters

Compensation varies dramatically across the three archetypes. Understanding the bands lets job-seekers target roles paying $160,000–$340,000 without competing in the most saturated applicant pools.

AI-related jobs pay approximately 28% more than comparable non-AI roles overall. But the premium is concentrated in technical archetypes. Governance roles, while growing 30–40% in demand, tend to cluster in the $160,000–$250,000 range at large enterprises and consultancies.

The strategic implication is clear. A software engineer pivoting to integrity engineering or a compliance professional upskilling in NIST AI RMF can access the $160,000–$340,000 band without needing the rare alignment research pedigree that red-team roles demand. The premium is real, but it is not evenly distributed, and the distribution follows the archetype map.

Where Each Archetype Lands

Each archetype maps to a distinct employer type. Understanding this distribution keeps job-seekers from applying to organizations that don't hire their profile.

Frontier labs (Anthropic, OpenAI, Google DeepMind, Thinking Machines Lab, Safe Superintelligence Inc., Cohere) concentrate red-teamers and integrity engineers. These labs poach aggressively during industry upheaval. When OpenAI researchers began tendering resignations in November 2023, Salesforce CEO Marc Benioff posted on X matching full cash and equity OTE for any of them who joined Einstein Trusted AI. On the same day, Cohere CEO Aidan Gomez posted a link to the company's career page seeking ML Members of Technical Staff. The talent wars are real, and they are fought along archetype lines.

Enterprise and government employers (the US federal government, BFSI firms, large consultancies) are the primary buyers of governance talent. The Horizon Fellowship and TechCongress serve as on-ramps into federal policy and legislative roles. NIST's April 2026 concept note signaled that federal demand for governance professionals would keep growing.

Job-seekers should target their archetype's employer cluster. A red-teamer applying to a Big Four consultancy is wasting cycles. A governance professional applying to Thinking Machines Lab is doing the same.

How to Break In Without Starting Over

Each archetype has identifiable on-ramps (certifications, fellowships, and skill adjacencies) that let professionals transition without returning to school or accepting entry-level positions.

For red-teamers, MATS and ARENA programs, plus open-source alignment research contributions, serve as credible signals. A background in security research, formal methods, or ML engineering transfers directly. The barrier is technical depth, not credentials.

For integrity engineers, MLOps experience, security engineering, and production ML monitoring skills are the core prerequisites. Certifications like CISO31K provide a salary-lift pathway, with alumni reporting a median 47% increase. The transition from production ML engineering to integrity engineering is a lateral move with a significant pay premium.

For governance professionals, NIST AI RMF fluency, policy fellowships, and domain expertise in regulated industries are the primary credentials. The 30–40% salary increases for compliance framework expertise reward targeted upskilling. A compliance professional who learns the NIST framework's four functions (Govern, Map, Measure, Manage) and can articulate how they integrate into CI/CD pipelines has a credential that hundreds of job postings now require.

The common thread: each archetype rewards a specific, demonstrable skill stack rather than a generic "AI safety" label. The $160,000–$340,000 band is accessible to those who build and signal the right stack.

The field's fragmentation into three distinct career tracks is not a sign of immaturity. It is the hallmark of a profession coming into its own — one where the job postings have already split, the salary bands have already diverged, and the only people still treating "AI safety" as a single job title are the ones losing money on every application.

Working in AI? Zero G Talent tracks the openings: browse AI jobs, openings at OpenAI and Anthropic, and the people building the field.