OpenAI partners with Hacktron to test third-party service connections – Unite.AI

OpenAI announced a partnership with Hacktron on November 6, 2026, to identify vulnerabilities in its systems, focusing on connections with third-party services and expanding the security company’s access to frontier models through Daybreak Red.

OpenAI engineers and Hacktron security researchers will jointly select high-priority areas to test. The work will examine how systems interact as AI agents connect services in new ways, with lessons used to strengthen OpenAI’s defenses. This gives the partnership a specific focus: weaknesses that emerge when applications, identity systems and infrastructure are connected.

A Bug Bounty relationship expands

OpenAI became familiar with Hacktron through its Security Bug Bounty program and said it was impressed by the impact of the vulnerabilities discovered and disclosed by the company’s researchers. Hacktron creates AI agents for code review, penetration testing, and red teaming, and its research has uncovered vulnerabilities in the interactions between applications, cloud infrastructure, and identity systems. These capabilities align directly with the new testing focus.

The Hacktron account from September 13, 2026 provides a concrete example. Its researchers said they chained two critical vulnerabilities on July 25, 2026, to compromise the ChatGPT accounts of multiple OpenAI employees, creating potential access to internal repositories and other connected services. The chain combined remote code execution in the image processing path of the Discourse-hosted community forum with an OpenAI single sign-on issue, allowing the forum compromise to progress to ChatGPT and Codex and then to a connected GitHub integration.

To demonstrate access without reading sensitive internal code, Hacktron used an employee’s Codex account to create a benign pull request in OpenAI’s internal monorepo, then stopped testing. He reported the issues via OpenAI’s Bugcrowd program and Discourse. Hacktron’s timeline states that OpenAI confirmed a fix about 14 hours after the initial report, while Discourse had a fix ready two days after receiving the report.

OpenAI paid Hacktron a $6,500 bounty on September 1, 2026, making it clear that the reward recognized OpenAI-side discovery rather than testing against the explicitly out-of-scope Discourse-hosted forum, according to the timeline reproduced by Hacktron.

Access and controls Daybreak Red

OpenAI’s August 10, 2026 Daybreak announcement defines Red as providing access to specially trained cybersecurity models for authorized vulnerability scanning, exploit validation, and security testing, and recommends Blue as the starting point for most defenders. Blue provides generic border models with adequate protections for authorized defensive work. Teams whose authorized work includes advanced vulnerability research, exploit development, or red teaming can request Red access.

Access to both levels is limited to approved individuals and organizations carrying out authorized work. Controls include identity verification, account security, monitoring, approved use restrictions, and legal attestations. Documented OpenAI best practices also call for isolating security workflows from sensitive production systems and the open internet, reviewing agent actions outside the Codex sandbox before execution, adding human oversight for higher-risk workflows, and using scoped authorization profiles to enforce authorized limits.

Previous Hacktron reporting also states that the work was not fully autonomous and that trained human guidance remained important. The defensive value of the partnership is therefore based on a balance: expanded access to cyber-capable models for more thorough testing of interconnected services, coupled with applicable reach, monitoring and human oversight when agents are used.