Reflecting on three of my favorite red teaming events from this past year

EVENT 1: Red teaming with NIST

Humane Intelligence, supported by the U.S. National Institute of Standards and Technology (NIST), conducted a nationwide AI red-teaming exercise to test and evaluate generative AI systems’ robustness, security, and ethical implications. This event was open to U.S.-based individuals and model developers, and the initiative advanced AI resilience and trustworthiness through rigorous adversarial testing and analysis. We sought to demonstrate capabilities to rigorously test and evaluate the robustness, security, and ethical implications of cutting-edge AI systems through adversarial testing and analysis. 

This exercise was crucial for helping to ensure the resilience and trustworthiness of AI technologies. We used the NIST AI Risk Management Framework 600-1 to explore generative AI risks and suggested actions as an approach for establishing generative AI safety and security controls. Together, we assessed violative outcomes and controls, working closely with developers to mitigate risks and refine these systems. This partnership showed that U.S. Federal Government agencies involved in setting safety standards cared deeply about engaging the public in the process of refining their standards. Further, we collaborated with four model owners – Anote, Meta, Robust Intelligence (now part of Cisco), and Synthesia – to identify vulnerabilities and improve the safety and trustworthiness of AI systems. Last, we tested a framework, showing that red teaming can be applied in diverse contexts. 

Humane Intelligence at NIST

EVENT 2: Red Teaming with IMDA Singapore

Humane Intelligence, in collaboration with IMDA Singapore and global leaders like Anthropic, AWS, Cohere, Google, and Meta, launched the first regional AI safety red teaming challenge in Asia. Participants from nine countries, including sociologists, linguists, and cultural experts, red teamed four large language models to spot biases and stereotypes directed at social groups in their countries, both in English and regional languages. This groundbreaking initiative addresses gaps in AI safety evaluations, which often focus on Western contexts, and seeks to develop a common methodology for safer, culturally sensitive AI models. This event served as the largest AI red teaming event of its kind in the Asia Pacific region, introducing people from across the region to collaborate on AI multilingual and multicultural evaluations. 

IMDA Singapore Red Teaming Participants

EVENT 3: Red teaming with UNESCO

Humane Intelligence partnered with UNESCO in Paris for a red-teaming exercise on Technology-Facilitated Gender-Based Violence (TFGBV). Held on the International Day for the Elimination of Violence Against Women and Girls, the event aimed to identify vulnerabilities in generative AI models and explore their potential misuse in perpetuating bias, discrimination, and violence against women and girls. Participants in this event—including senior diplomats and UNESCO experts—examined embedded biases and assessed whether safety measures in generative AI models hold up under scrutiny. 

This collaboration builds on UNESCO’s extensive work on gender and AI, including their recommendations for addressing bias and their global reports on AI-facilitated gender-based violence. By translating theoretical insights into practical solutions, this partnership reinforces efforts to ensure AI systems prioritize fairness, safety, and inclusivity, which is at the core of Humane Intelligence’s mission. This event introduced UNESCO’s staff, who work to foster international cooperation in education, science, culture, and communication to promote peace and sustainable development, to the opportunities and challenges that generative AI presents, in the context of TFGBV, an issue that affects many of us personally. 

Humane Intelligence at UNESCO HQ in Paris, France. (c) UNESCO/Capucine Petit