OpenAI Launches Bug Bounty Program to Test Claude 3.7 Safety Defenses

AI Impact Summary

This announcement details a new bug bounty program focused on testing the safety defenses of Claude 3.7 Sonnet, specifically targeting universal jailbreaks in Constitutional Classifiers. The program is a proactive measure to validate the effectiveness of ASL-3 safeguards, which are crucial for mitigating risks associated with increasingly capable AI models, particularly those related to CBRN weapons. This initiative represents a key step in our Responsible Scaling Policy and demonstrates our commitment to responsible AI development.