Clinical AI Safety Evaluator

Overview Join a select team of healthcare experts focused on safeguarding and refining AI systems. This role centers on evaluating clinical precision, safety standards, and overall response quality in generative healthcare AI applications. Key responsibilities involve performing structured evaluations using established clinical frameworks and conducting adversarial stress-testing (Red-Teaming) to identify potential vulnerabilities and safety concerns before public release.

We are looking for experts operating at the intersection of clinical practice, AI expertise, and health informatics safety. Candidates should be practicing clinicians passionate about health technology, possessing the analytical rigor needed to push model limits and protect patient safety

This is a vendor role with Highspring for a client’s team, a Fortune 100 company. This position is remote until further notice.

Responsibilities: Structured Grading

* Reviewing prompt-response pairs and scoring them across dimensions like clinical accuracy, safety, and various health risk domains.

Dynamic Red-Teaming Sprints

* Participating in “sweeps” where they dynamically converse with the model to probe specific health risks (e.g., attempting to get the model to prescribe medication or diagnose a complex condition).

Providing Clinical Rationale

* Writing clear, concise, and scientifically backed justifications for why a model response was graded as unsafe or inaccurate.

Minimum Qualifications Clinical Credentials * MD, MBBS, or local equivalent from an accredited institution.

* Candidates must be Board-Certified in one of the following priority specialties: ○ Psychiatry / Clinical Psychology (Mental health, self-harm risks)

○ Primary Care / Family Medicine / Internal Medicine (General health queries)

* Active, unrestricted clinical licensure in good standing in their practicing geography (US).

* Direct, current practice in patient care. Understanding how patients, carers and families ask questions in the real world is critical for realistic testing.

AI Literacy * Active experience using generative AI platforms (e.g., Gemini, ChatGPT, Claude) for clinical workflows, research, or everyday queries.

Red Teaming Clinician Rater

* Understanding of the inherent limitations of LLMs, including hallucinations, sycophancy (the model agreeing with the user even when wrong), bias, and the risks of generating unsafe clinical advice.

* Ability to write clear, structured, and complex prompts to interact with models dynamically

Evaluations & Red-Teaming * Ability to adopt different user personas (e.g., a worried parent, a patient seeking to bypass safety rules, an anxious patient misinterpreting symptoms) to stress-test the model.

* Experience working with complex, multi-dimensional rubrics (such as Likert scales). Comfort with aligning their grading to strict “Anchor Examples” to minimize subjective bias.

Preferred Qualifications * An additional advanced degree in a relevant field (e.g., Master of Public Health (MPH), MS/PhD in Medical Informatics, Computer Science, or Behavioral Science).

* Fluency in multiple languages and deep understanding of cultural health beliefs and regional clinical guidelines (vital for global scaling).

* Previous experience designing, implementing, or participating in AI safety evaluations, Trust & Safety workflows, or RLHF (Reinforcement Learning from Human Feedback) for medical models.

* Debarment and Exclusion. Contractor represents and warrants that it has not been and will not use in any capacity any individual or entity to perform any Services if such individual or entity has been:

(i) debarred by the Food and Drug Administration (“FDA”) pursuant to 21 U.S.C. § 335a; (ii) excluded from participation in any state or federal healthcare program, as defined in 42 U.S.C.§1320a-7b(f); (iii) debarred from participation in federal procurement and/or non-procurement programs; (iv) on FDA’s list of investigators with disqualification proceedings, or otherwise deemed ineligible to receive investigational products; (v) charged, or convicted of a criminal offense, or named or found liable in an action under any applicable law pertaining to the regulation of a health related product or the provision of health related services; (vi) the subject of a final or pending adverse action, as defined in 42 U.S.C. § 1320a-7a(e) and 42…