In July 2026 OpenAI was in the news regarding its AI models performing a cybersecurity test on an experimental AI model inside a “digital sandbox.” The AI agents bypassed the sandbox, accessed the internet, and ultimately hacked AI research platform, Hugging Face. OpenAI’s response to the matter was conciliatory. The incident glaringly illustrated how vulnerable AI systems can become.
OpenAI’s recent collaboration with Hacktron addresses its AI models potential security concerns. OpenAI recently announced its partnership with Hacktron to assess vulnerabilities in it’s systems. OpenAI engineers and Hacktron’s security researchers will address high-priority areas to test together, and on how its systems interact with third-party services.
OpenAI is setting a strong example by putting frontier cyber models in the hands of security researchers, enabling them to uncover complex vulnerabilities at the boundaries between interconnected systems.
We’re excited to deepen our partnership with OpenAI, combining our offensive security expertise with their frontier models to help secure OpenAI and the broader software ecosystem.
—Zayne Zhang, CEO, Hacktron.
OpenAI will also give Hacktron’s access to its frontier models via Daybreak Red.
Through OpenAI’s Security Bug Bounty program, Hacktron researchers found vulnerabilities in its AI processes:
The team builds AI agents for code review, penetration testing, and red teaming. Their research has uncovered vulnerabilities in how applications, cloud infrastructure, and identity systems interact.
Hacktron’s expertise will analyze AI agents when connecting to services in novel ways. OpenAI will glean valuable insight from this process:
Working alongside Hacktron will help us test those connections more deeply and use what we learn to strengthen our defenses.