Strategic Projects Lead, Safety

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

Runs adversarial red-teaming and safety evaluation programs that push frontier AI models to failure before and after release, covering sensitive, high-severity harm categories. The team works directly with frontier AI labs to turn findings into sharper policies and broader evaluation coverage.

• AI misuse and jailbreaks

What you'd do

• Own end-to-end delivery of red-teaming and safety evaluation programs, from scoping through completion.
• Manage several programs for different frontier labs, each with separate harm categories and contacts.
• Design adversarial testing methods such as jailbreak development and repeated testing that pushes models to failure.
• Extend red-teaming into agentic settings where models use tools and multi-step autonomy under adversarial pressure.
• Lead and coordinate expert Fellows doing sensitive testing across harm categories, languages, and markets.
• Work with lab policy leads to spot coverage holes, unclear labels, and new risk areas.
• Synthesize testing data into insight reports on trends, systemic gaps, and recommended policy and evaluation work.

What it asks for

• Two-plus years in T&S, red-teaming, security research, policy enforcement, or an adjacent technical or analytical field.
• Strong project and workforce management skills, including leading teams of testers on tight timelines.
• Analytical, first-principles problem-solving in ambiguous, fast-changing testing environments.
• Working knowledge of jailbreaks and prompt exploits, or ability to pick them up quickly.
• Strong communication and stakeholder handling, including with senior customers and lab policy staff.
• Curiosity and ability to learn AI concepts, model behavior, and industry trends quickly.
• Experience red-teaming or adversarially evaluating LLMs or agentic AI systems (preferred).

Worth knowing

• Exposure to sensitive or explicit content, including violent, sexual, or otherwise disturbing material.
• Real-time decisions affecting delivery quality and program scope with frontier labs.

Terms in this role

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.

Jailbreak.

A prompt written to get an AI model to ignore its safety rules.

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