What to know
About Handshake AI
Handshake AI is the AI training division of Handshake. The parent company was founded in 2014 by Garrett Lord, Ben Christensen and Scott Ringwelski and is based in San Francisco, California. Handshake originally became known as a career platform for students and graduates. It formally introduced Handshake AI in June 2025 as a service connecting skilled people with AI companies that need human expertise to train and test their systems.
Through the Handshake AI Fellowship, you can work on projects for leading AI labs. Your role is to use your knowledge and judgment to identify mistakes, improve answers and help AI systems perform more reliably.
AI data training work at Handshake AI
Typical tasks include reviewing and editing AI-generated content, labeling or organizing information, suggesting better responses and testing how an AI model handles real-world situations. An AI model is simply the system behind tools such as chatbots and other generative AI products.
Handshake recruits both generalists and subject specialists. Current opportunity categories include computer science, engineering, biology, medicine, law, finance, business, mathematics, psychology, music, design, performing arts and the humanities. Some entry-level projects accept applicants from any academic subject and do not require previous AI training experience. More specialized projects may require a particular degree, professional background or advanced subject knowledge.
What is it like to work at Handshake AI?
Handshake AI Fellows work as independent contractors rather than employees. Projects are remote, part-time and generally completed independently. There is no fixed schedule, minimum number of hours or guarantee of weekly work. You can usually choose when to work, but tasks may appear or disappear quickly depending on demand from the partner AI lab.
Project length varies. Handshake says an invitation normally includes the estimated duration and any expected time commitment. Being accepted into the Fellowship also does not guarantee a project. Your profile enters a matching pool, and invitations depend on your background and project demand.
Most projects pay either an hourly rate or a fixed amount per completed task. Public examples currently include up to $30 per hour for a generalist trainer and up to $125 per hour for certain specialist roles. “Up to” is important: your actual rate depends on the specific project terms. Onboarding may be paid or unpaid, optional meetings are usually unpaid and some tasks have limits on how much time can be billed. Payments are processed through Stripe, typically on a weekly cycle, although some projects may use a different schedule.
Recent research
Handshake’s 2026 research includes CAREBench, a benchmark for upstream child-safety risks in language models, and Gandalf, a verifier that grades agent work inside the same tool environment used to complete it. These projects illustrate why some Fellowship tasks involve building realistic tests, checking tool use and judging complete work products rather than only rating short text answers.
Application process at Handshake AI
You start with a Fellowship or opportunity application and a complete Handshake profile. Handshake’s June 2026 application guidance says there is no fixed review timeline. If your experience fits a project, an interest email explains the work, estimated duration, expected commitment, pay and requirements before you decide whether to continue.
Project selection is competitive and based on fit, available capacity and prior performance. Selected Fellows complete project-specific onboarding, which can include training material, a knowledge quiz, practice task, written test or skills assessment. Passing opens that project when task capacity is available; failing one assessment does not necessarily remove your broader Fellowship profile. The Fellowship expectations also distinguish required project work from optional meetings and other nonbillable activity.
Sources: Company website · Terms · Privacy · Handshake AI research
Information last updated August 2026.