Senior Freelance Consultant, AI Safety

Written by the T&S Network from Moonshot's posting.

Red teaming and adversarial evaluation of AI systems across harm categories including violent extremism, pathways to violence, CSEA and grooming, mental health and crisis, and risks to children and teens. The work sits in Moonshot's AI Safety portfolio, where the consultant supports the team in building evaluation frameworks, test scenarios and scoring criteria.

• Child sexual abuse material
• Child safety and grooming
• Terrorism and violent extremism
• Self harm and suicide

What you'd do

• Stress-test AI systems adversarially across defined harm categories.
• Review model responses against harm and risk criteria and give expert judgement on them.
• Apply subject matter expertise to a specific harm area such as grooming, radicalisation, crisis signalling or teen online safety.
• Co-build evaluation frameworks that convert real-world harm expertise into criteria that can be tested.
• Contribute to intervention logic that connects at-risk users to appropriate support.
• Draft methodology and findings for technical and government audiences.

What it asks for

• Trust and safety, online harms, or adjacent work such as violence prevention or safeguarding, applied to AI systems.
• Proven record building research, evaluation frameworks or interventions on extremism, CSEA, self-harm or targeted violence.
• Skill in converting how a harm operates into tests that show whether an AI system responds safely.
• Strong written communication for technical and government readers.
• Availability for close to full-time work over about six weeks.
• Willingness to complete security clearance procedures if the engagement requires them.
• Depth in AI model safety or red teaming of LLMs (preferred).

Worth knowing

• Exposure to graphic, violent, and extremist content as part of the work.
• Freelance consultancy on a project basis, starting as soon as possible.
• Required security clearance if the engagement calls for it.

Terms in this role

CSEA.

Child sexual exploitation and abuse. The wider harm area around CSAM, including grooming, sextortion and trafficking.

Red teaming.

Testing a system, often an AI model, by deliberately trying to make it fail or misbehave, so problems are found and fixed first.

Our best summary of the posting, and it can be wrong. The employer's posting is what counts: read it in full on Moonshotnorth_east before you apply. If we got something wrong, report this listing.