Unlike traditional bug bounties that target code errors, Humane Intelligence’s algorithmic bias bounties focus on discovering the root causes of biased or exclusionary outcomes in AI systems. Instead of treating bias as an abstract or philosophical debate, bias bounties create a structured process where bias can be systematically surfaced, measured, and addressed.
Thanks to the support of the Heising-Simons Foundation and in partnership with Radiant Earth and Reliabl, as of June 2026, Humane Intelligence has moved our bias bounty program over to Zindi, a global data science platform with users in more than 185 countries! This helps us lower costs and reach more participants. Please reach if you’d like a referral to the Zindi team or our help to scope and launch a challenge.
Humane Intelligence takes a hands-on approach to ensure every bounty is impactful, well-executed, and aligned with our partners’ goals. We combine expertise in bias, sociotechnical research, and data science, and work closely with our organizational partners to co-design each challenge scope, engage the right participants, and evaluate findings in a way that honors impacted communities while also driving technical improvement.
Participants use systematic testing methods to uncover issues like biased training data, discriminatory default settings, and algorithmic blind spots that fail to account for human diversity. Beyond documenting exclusionary patterns, participants also design and develop technical solutions that enhance system performance in real-world conditions.
SPECIAL THANKS
Our first ten challenges were launched
Humane Intelligence partnered with Valence AI and CoNA Lab on a bias bounty challenge focused on accessibility for neurodivergent people in conferencing platforms like Zoom, and on the role of emotion AI detection in shaping those experiences. Participants will be able to choose from a design or machine learning track to identify accessibility gaps and propose improvements.
The challenge dates were:
Design Track
Data Track
Humane Intelligence partnered with Indian Forest Service for this challenge set. In three levels and tracks – thought leadership, beginning technical, intermediate technical – focused on ensuring fair, biophysically informed, and community-driven tree planting site recommendations—tackling bias in AI-driven environmental decision-making.
The challenge dates were:
Humane Intelligence partnered with Revontulet for this challenge set. In two levels – intermediate and advanced – participants focused on counterterrorism in computer vision (CV) applications, centered on far-right extremist groups in Europe / the Nordic region. The goal was to train a CV model to understand the ways in which hateful image-propaganda can be disguised and manipulated to evade detection on social media platforms.
The challenge dates were:
In three levels – beginner, intermediate, advanced – participants designed fine-tune automated red teaming models to explore issues like bias, factuality, and misdirection in Generative AI.
The challenge dates were:
