Moonshot is seeking a Senior Freelance Consultant to support the delivery of our AI Safety portfolio. This is a hands on, technical engagement focused on red teaming, adversarial evaluation, and methodological development across harm categories including pathways to violence, extremism, child sexual exploitation, abuse and grooming (CSEA), mental health and crisis, and risks affecting children and teenagers.
The consultant will work closely with Moonshot 's AI Safety team to design and execute evaluation frameworks, test scenarios, and scoring criteria, translating violence prevention, safeguarding, and behavioural risk expertise into structured, testable outputs suitable for technical and policy audiences.
This role does not include client relationship management, team leadership, or business development responsibilities. The focus is on the technical and methodological work itself, and all work must be delivered within a clear ethical framework, in compliance with contractual, legal, data protection, and ethics obligations.
What the work involves
Engagements vary, but may involve:
Red teaming and adversarial evaluation of AI systems against defined harm categories.
Reviewing model responses against harm and risk criteria and providing expert judgement.
Bringing subject matter expertise to a specific harm area, such as grooming and CSEA, radicalisation pathways, crisis signalling, or teen online safety.
Supporting the design of evaluation frameworks that translate real world harm knowledge into structured, testable criteria.
Contributing to the design of intervention logic that connects at risk users to appropriate support.
Drafting methodology or findings suitable for technical and government audiences.
Requirements
Essential:
Experience in trust and safety, online harms, or a closely related field such as violence prevention, safeguarding, or public health, with the ability to apply that knowledge to AI systems.
Demonstrated experience designing research, evaluation frameworks, or interventions for harm categories such as violent extremism, CSEA, self-harm and crisis, or targeted violence.
The ability to translate real world knowledge of how a harm works into a way of testing whether an AI system handles it safely.
Comfort and demonstrated resilience working with highly sensitive or graphic content (violence, extremist material, crisis content), with awareness of wellbeing practices for this kind of work.
Strong written communication, able to produce credible, non promotional material for technical and government audiences.
Sound judgement working with ambiguity and sensitive material.
Availability for a close to full time commitment over approximately 6 weeks.
Willingness to undertake relevant security clearance procedures if required by the engagement.
Desirable, genuine depth in one or more of the following:
Model safety, red teaming, or adversarial evaluation of LLMs or other AI systems.
Understanding of LLM architecture, safety tooling, or trust and safety policy.
Child safety evaluation, teen safety product work, or grooming and CSEA detection.
Intervention or diversion programme design that can transfer to AI mediated interventions.
Government or regulatory engagement, such as briefing officials or supporting policy submissions.
An academic or applied background in radicalisation studies, forensic psychology, or violence risk assessment.
Taxonomy or classifier development, including how testing data feeds a classifier.
Benefits
As a consultant, you will have:
Flexible working arrangements.
Opportunity to work on diverse, impactful projects.
Competitive consultancy rates.
Remote working options available.
We welcome expressions of interest across a range of experience levels, from established specialists to people earlier in their careers who bring real depth in a relevant area.
Send us your CV and a short statement of interest, no more than one page. Your statement should set out the experience you'd bring from the areas above, and the harm categories or methods you have the most depth in.
