Surge AI is a data labeling and AI training data company that provides high-quality human annotation, RLHF datasets, and adversarial red-teaming services to frontier AI labs including Anthropic, OpenAI, Google, Microsoft, and Meta.
Surge AI is a data labeling and AI training data company that provides high-quality human annotation, RLHF datasets, and adversarial red-teaming services to frontier AI labs including Anthropic, OpenAI, Google, Microsoft, and Meta.
People
Updated 04/02/26Founder & CEO
Strategic Partnerships
Recruiter
Advisor
Design
Strategic Partnerships
Strategic Partnerships
Growth Marketing
Funding Details
Updated 04/02/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- -
Org Details
Updated 04/02/26Surge AI was founded in May 2020 by Edwin Chen, a former research scientist at Google, Facebook, and Twitter, and MIT alumnus in mathematics, computer science, and linguistics. Chen started the company out of dissatisfaction with the low quality of crowdsourced data labeling platforms, aiming to build a marketplace of expert human annotators who could produce training data that captures the full richness of human intelligence and judgment. The company's core offering is a platform that converts raw text, code, images, and conversation transcripts into structured training data used to train large language models. Surge specializes in reinforcement learning from human feedback (RLHF), policy-aligned safety datasets, model evaluations, and RL environment construction. Their annotator network consists of approximately 50,000 expert contractors globally, carefully vetted for domain expertise. Surge serves approximately 12 frontier AI labs, with major clients including OpenAI, Anthropic, Google, Microsoft, and Meta. The company played a notable role in improving Claude's capabilities, particularly for coding tasks. Beyond standard annotation, Surge has partnered with AI safety organizations such as Redwood Research to run adversarial red-team labeling campaigns, in which human labelers generate examples designed to probe and break AI classifiers, enabling labs to build more robust and reliable safety systems. Surge maintained a lean operating model throughout its history. With fewer than 130 full-time employees, it surpassed $1 billion in annual revenue in 2024, making it one of the most capital-efficient companies in the AI industry. The company was entirely bootstrapped and profitable from its earliest days. In July 2025, Surge initiated its first external fundraising round, reportedly seeking approximately $1 billion at a valuation between $15 and $25 billion. In addition to labeling services, Surge publishes evaluation benchmarks such as Hemingway-bench (for AI writing quality) and EnterpriseBench: CoreCraft (for AI agents in enterprise environments), reflecting the company's broader commitment to rigorous, real-world AI evaluation.
Theory of Change
Updated 04/02/26Surge AI's theory of change holds that the quality of human-generated training data is the primary determinant of whether AI systems are safe, aligned, and genuinely useful. By supplying frontier AI labs with expert-annotated RLHF datasets, safety-policy corpora, and adversarial red-team examples, Surge helps shape the values and behaviors embedded in deployed models. Their adversarial labeling work—generating examples designed to fool classifiers—directly supports AI safety research by enabling labs to build more robust, failure-resistant systems. The underlying premise is that human judgment and taste, injected at scale through careful data curation, is the critical input that determines whether AI development trends toward beneficial or harmful outcomes.
Grants Received
Updated 04/02/26Projects– no linked projects
Updated 04/02/26Discussion
No comments yet. Be the first to share your thoughts.