OpenAI is partnering with the security firm Cathedral to identify vulnerabilities within its systems. Researchers from both organizations will use OpenAI’s AI models to test the company’s defense mechanisms and evaluate the effectiveness of potential fixes. Furthermore, OpenAI has more than quadrupled the rewards offered for critical vulnerabilities reported through its Security Bug Bounty program. This initiative follows a previous round of security testing, during which OpenAI tasked 25% of its product engineers with identifying and patching vulnerabilities.
Cathedral researchers will work alongside the OpenAI team.
In a blog post, Cathedral stated that its researchers will join forces with OpenAI’s team to uncover security vulnerabilities in the company’s systems. They will leverage OpenAI’s AI models to stress-test defense mechanisms and verify the efficacy of proposed solutions. Gavin Kliger, CEO and co-founder of Cathedral, noted that this collaboration will enable his firm’s security researchers to work hand-in-hand with the OpenAI team to detect weaknesses that might otherwise go unnoticed.
OpenAI has also increased the rewards offered through its bug bounty program to researchers who report critical security vulnerabilities. While the company has not disclosed the specific new reward amounts, it stated that payouts for key findings will be more than four times higher than before.
This new initiative builds upon a previous security effort, during which OpenAI tasked 25% of its product engineers with identifying and patching vulnerabilities. They utilized Codex to investigate reported issues, reproduce them within isolated environments, and prepare fixes for review.
OpenAI is also granting access to its AI models to Daybreak’s clients. The company has pledged $1 billion (approximately 95 billion rupees) in subsidies over the next six months to help organizations that protect critical services and digital infrastructure leverage artificial intelligence in their cybersecurity efforts.
The company plans to share lessons learned from its recent security exercise, including how it was organized, as well as the tools and processes employed. Furthermore, the company will release a guide to serve as a resource for other companies and government agencies looking to leverage artificial intelligence to enhance their cybersecurity.