Harvard in Tech AI Edition: Leonard Tang, co-founder and CEO of Haize Labs
Leonard Tang, co-founder and CEO of Haize Labs, is on a mission to build the trust, safety, and reliability layer for AI. His journey to founding Haize Labs is both unconventional and inspiring. Initially, Leonard was deeply skeptical of the startup world during his time at Harvard, considering it a distraction from his passion for machine learning research — much of which now forms the foundation of Haize Labs. He had planned to pursue a PhD at Stanford, set to begin in September, but ultimately chose to dedicate himself fully to the company.
Leonard’s early experiences in startups, including an attempt at Y Combinator with a hardware-focused functional verification idea, marked a pivotal moment in his career. Despite being accepted into the prestigious accelerator, he initially resisted dropping out of school. It wasn’t until he observed the rise of overly hyped AI companies — many of which he felt were deploying unreliable “demoware” — that he saw a pressing need for a company like Haize Labs. Convinced that existing approaches to AI safety and evaluations were inadequate, he made the decision to forgo his PhD and focus on building a solution that could meet the challenges of AI deployment at scale.
Leonard draws parallels between his research background and his journey as a founder. Both involve navigating uncharted territory without a clear roadmap. The process of making informed decisions step by step is something Leonard thrives on. His experience with Gamalon, the first company where he interned, was particularly formative. Under the mentorship of its CEO, a pioneer in probabilistic inference chips, Leonard gained firsthand insight into how market needs shape research and how research informs product development. This fusion of academic rigor and practical application is something he continues to emphasize at Haize Labs.
Haize Labs maintains a strong focus on research, which Leonard identifies as its competitive edge. Unlike competitors that rush to market with undifferentiated SaaS solutions, Haize Labs takes a deliberate approach, aiming to deliver meaningful innovation. Drawing from their expertise in adversarial attacks, synthetic data generation, scalable oversight, and active learning, the company is developing high-quality, low-latency automated evaluation tools that do not require the extensive budgets of foundation model companies.
One area Haize Labs has recently explored is AI for content moderation, a growing challenge as generative AI systems produce more content than humans can feasibly review. Leonard points out that while existing solutions, like Hive, perform well in moderating images and videos, text-based content presents a far more complex problem. The infinite variability of language, combined with differing cultural sensitivities and thresholds for acceptability, makes text moderation technically and contextually challenging.
Haize Labs’ mission resonates with diverse stakeholders, uniting those focused on AI safety with those seeking to enhance AI capabilities. Leonard emphasizes that safety is fundamentally about trust — ensuring that AI systems reliably perform as expected. He also notes the importance of localized approaches to AI safety, as different countries prioritize unique cultural norms and safety concerns.
As the industry advances, Leonard foresees growing challenges with the rise of AI agents, which amplify reliability issues by chaining multiple model calls together, compounding error rates. These agents, especially when given autonomy, introduce new risks — such as the potential for unmoderated actions that can directly influence human beliefs.
When asked for advice for students at Harvard, Leonard encourages them to seek out inspiring people, leverage the university’s name to connect with experts, and not shy away from reaching out. Reflecting on his own experience, he notes that much of his research was conducted at institutions like MIT and Berkeley due to the relatively nascent state of the machine learning community at Harvard during his time there. His parting wisdom: surround yourself with people who challenge and inspire you.
