Behavioral Health Expert, AI Safety and Model Evaluation

What you'll do

People now ask AI about their marriages, their families, their faith and their worst decisions. Most of these conversations are not crises, which is exactly why a model can get them wrong quietly: by agreeing too readily, taking a side, validating a distorted belief, lecturing, or drifting into clinical advice it has no business giving. A leading AI lab wants behavioural health professionals to judge where that line sits.

The work:

• Evaluating conversations on relationship and family advice, emotional wellbeing, spiritual and metaphysical questions, and unconventional or unfounded beliefs, judging whether responses are neutral, appropriate and safe
• Naming the patterns (excessive agreement, taking sides, reinforcing unfounded beliefs, moralising, overstepping into directive advice) with clear written rationale
• Writing the standard: rubrics, guidelines and reference responses for a balanced, supportive, bounded reply, grounded in counselling and behavioural health practice
• Designing test scenarios that probe sensitive but non-crisis topics
• Calibrating with the lab's researchers and other experts

For therapists, this is close to supervising a trainee's session notes: you already know the difference between empathic reflection and collusion. The skill to build is explaining that difference in a sentence a researcher can turn into a rule.

Who fits

• A degree in psychology, counselling, social work, behavioural health, behavioural science, human services or a related field, or equivalent professional mental health experience
• 3+ years supporting people in a mental health, counselling or social services setting (therapist, counsellor, clinical social worker, psychologist, psychiatric nurse, case manager, crisis counsellor, peer support specialist or advocate)
• Able to stay neutral and nonjudgmental across perspectives and worldviews, and explain your reasoning
• Working familiarity with sycophancy, cognitive distortions, healthy boundaries and client-centred approaches such as motivational interviewing
• At least 20 hours a week on weekdays
• Strong, precise written feedback
• US-based

Licensure (LMFT, LCSW, LPC, LMHC, PsyD, PhD) is valued but not required, which opens this to case managers and peer support specialists. Nice to have: AI safety, applied ethics, trust and safety or content policy background; couples or family counselling; familiarity with spiritual, religious or alternative belief communities, or the psychology of misinformation; prior AI evaluation or red teaming.

What it pays

$45–70 per hour. At 20 to 40 hours a week that is roughly $900 to $2,800 weekly before tax. The ad does not say what sets your rate; licensure is the obvious candidate. Mercor's clinical mental health conversation evaluation role is listed at $100/hour, so licensed clinicians should check whether that one fits them better.

Employment terms

This is a W-2 employment position with Cincinnatus LLC, an employer of record that handles payroll, benefits and compliance and places you within the lab's extended workforce. That makes it different from most Mercor listings, which are independent contractor roles: as a W-2 employee, tax is withheld from your pay rather than left for you to handle.

Worth knowing

Good:

• Licensure not required; 3 years of hands-on mental health work is the real bar
• W-2 employment through Cincinnatus, unlike most Mercor contractor roles
• Room to scale from 20 to 40 hours a week
• Rubric and scenario design, not just rating

Less good:

• Pay is modest next to clinical-licence evaluation roles
• 20-hour weekday minimum
• US only
• Reading many emotionally loaded conversations takes a toll; the ad says nothing about wellbeing support

About this listing

Posted by Mercor as a part-time remote role in the United States, for an unnamed leading AI lab, with Cincinnatus LLC named as the W-2 employer. The $45–70 band, the hours and the qualifications are the ad's own. See Mercor.