OpenAI partners with Cathedral Systems to hunt for AI security flaws

A proactive defense strategy

OpenAI is launching a major security initiative to identify and repair vulnerabilities in its systems. The company is teaming up with Cathedral Systems, a cybersecurity startup, to conduct a new internal security sprint. This move is designed to stay ahead of cyber threats that are becoming more advanced as artificial intelligence grows smarter.

The primary goal of this partnership is to use AI itself to strengthen the company's digital walls. OpenAI views this as a continuous effort rather than a one-time project. By using model-assisted testing, the company aims to make its systems more resilient against potential attacks from bad actors.

This strategy acknowledges a growing reality in the tech world. As AI models become more powerful, protecting internal secrets and user data becomes a top priority. OpenAI wants to ensure its digital defenses are just as sophisticated as the AI models it builds for the public.

The Cathedral Systems edge

Cathedral Systems is a relatively new player in the cybersecurity space, founded by former DOGE staffers. They have already demonstrated their capabilities by using AI agents to find security gaps in a rival firm's third-party business systems. That specific operation was completed in under 12 hours.

By bringing Cathedral Systems into the fold, OpenAI hopes to leverage these specific hacking methods. The startup is known for its fast and effective approach to finding hidden weaknesses. This collaboration allows OpenAI to see its own network through the eyes of a specialized security firm.

The partnership will allow external researchers to work alongside internal teams. These researchers will be able to use OpenAI's own models to challenge current defenses. This creates a feedback loop where the tools being developed are also used to stress-test the infrastructure they live on.

Boosting the bug bounty

To ensure the success of this security sprint, OpenAI is significantly increasing its bug bounty rewards. A bug bounty is a program where a company pays independent researchers to find and report software flaws. The company is now offering a payout of up to $172,000 for exceptional discoveries, including bonuses.

This new reward structure represents a massive jump in incentives for the security community. The updated amount is more than four times the previous reward for finding critical issues. By offering such a high payout, OpenAI is attempting to attract the best talent to help secure its software.

The company shared these updates on X, highlighting the importance of using AI for digital defense. They noted that because AI is getting better at finding vulnerabilities, the good guys must use it to secure systems first. This bounty hike is a key part of that broader strategy to stay ahead of hackers.

Lessons from previous sprints

OpenAI is not new to these types of intensive security exercises. During a previous security sprint, the company took a very aggressive approach to its internal audits. It dedicated 25 percent of its product engineers specifically to the task of finding and fixing security weaknesses.

During that previous exercise, engineers used Codex to analyze findings and reproduce vulnerabilities. They worked in safe test environments to ensure that the testing did not disrupt live systems. Once a flaw was identified, they created software patches to fix the issue immediately.

The company emphasizes that these sprints are part of an ongoing process to keep its systems safe. Each sprint provides data that helps the engineering team refine their defenses. By integrating these lessons into their daily operations, OpenAI aims to create a more robust environment for its users.

The future of AI security

The collaboration between OpenAI and Cathedral Systems signals a shift toward AI-driven security. As hackers begin to use automated tools to find bugs, companies must automate their own defense mechanisms. This