OpenAI brings Hacktron in to test its own systems, and expands that firm’s Daybreak Red access
Friday’s post says engineers from both sides will choose the targets, with the focus on how OpenAI’s systems meet third-party services. It does not name a price, a start date, or which models the access includes. OpenAI’s Daybreak help page says GPT-6 Cyber still requires a separate approval inside Red.
Short1 minEasy
Priya Shah1 min read
- OpenAI
- Hacktron
- Security
- Daybreak

Photo: OpenAISource page
OpenAI on Friday said it is partnering with Hacktron to look for vulnerabilities in OpenAI’s own systems. The post is dated 6 November. It describes a testing arrangement, not a newly found flaw. OpenAI’s engineers and Hacktron’s researchers will choose the high-priority areas together. The focus, the post says, is how OpenAI’s systems interact with third-party services. As part of the arrangement, OpenAI is expanding Hacktron’s access to frontier models through Daybreak Red.
OpenAI says it came to know the firm through its security bug bounty, and that it was impressed by vulnerabilities those researchers had found and reported. It describes Hacktron as a builder of agents for code review, penetration testing and red teaming. Zayne Zhang, Hacktron’s chief executive, is quoted saying OpenAI is “putting frontier cyber models in the hands of security researchers” so they can find weaknesses “at the boundaries between interconnected systems.” Hacktron’s own note, also dated 6 November and signed by Zhang, restates the same scope: third-party connections, and broader model access through Daybreak Red. Neither post names a fee, a term, or a date when the joint testing starts.
The OpenAI post points at two earlier moves in the same week. Cathedral’s researchers, it says, will work with OpenAI’s teams on the next security sprint, and bug-bounty rewards have been raised. It does not say the Hacktron work is part of that sprint. It also does not name a model. OpenAI’s current Daybreak help page says Red is limited to approved business and enterprise organizations, that those organizations are eligible for GPT-5.6 Cyber, and that GPT-6 Cyber needs a separate model approval. Turning Red on, the page says, does not grant that approval. The Hacktron announcement does not say the separate approval was given.


