OpenAI-Hacktron Security Partnership Expands Third-Party Testing

[jiaolix]

On November 6, 2026, OpenAI announced a partnership with Hacktron to identify vulnerabilities in its systems, specifically those related to connections with third-party services. Under the collaboration, Hacktron will receive expanded access to OpenAI’s frontier models through the “Daybreak Red” program. OpenAI engineers and Hacktron security researchers will jointly select high-priority areas for testing and examine how systems interact when AI agents connect to services in novel ways. The findings will be used to strengthen OpenAI’s defenses at the intersection of applications, identity systems, and infrastructure.

A Security Partnership Built on Bug Bounty Research

OpenAI first encountered Hacktron through its bug bounty program. OpenAI noted that the vulnerabilities discovered and disclosed by Hacktron researchers were particularly impactful. Hacktron develops AI agents for code review, penetration testing, and red-teaming exercises. Its research has uncovered flaws in how applications, cloud infrastructure, and identity systems interact—capabilities that align closely with OpenAI’s new testing priorities.

The partnership reflects a broader shift in AI security: defenders increasingly need to test not only individual models, but also the connections between models, identity systems, applications, and third-party services. As AI agents take on more tasks, those connections become more complex and more important to secure.

How Researchers Pivoted from Discourse to ChatGPT

A Hacktron report published on September 13, 2026, provided a concrete example of the risks involved. Researchers said that on July 25, 2026, they exploited two critical vulnerabilities to compromise the ChatGPT accounts of several OpenAI employees, potentially granting access to internal codebases and other connected services.

The exploit chain combined a remote code execution vulnerability in the image-processing path of the Discourse community forum with a flaw in OpenAI’s single sign-on system. This allowed the researchers to pivot from the forum to ChatGPT and Codex, ultimately gaining access to an integrated GitHub account.

To demonstrate access without reading sensitive internal code, the team used an employee’s Codex account to create a benign pull request within OpenAI’s internal monolithic codebase, then stopped testing. They reported the issues through OpenAI’s Bugcrowd program and directly to Discourse. According to Hacktron’s timeline, OpenAI confirmed a fix about 14 hours after receiving the initial report, while Discourse had a fix ready two days after being notified. OpenAI paid a $6,500 bounty on September 1, 2026, and clarified that the reward recognized discoveries made by the OpenAI team itself, not testing of the Discourse forum, which was explicitly out of scope.

Daybreak Red and Blue Access Tiers

OpenAI announced Daybreak on August 10, 2026. The Red tier provides access to a specialized, cybersecurity-trained model designed for authorized vulnerability research, exploit verification, and security testing. OpenAI recommended that most defenders begin with the Blue tier, which offers cutting-edge, general-purpose models equipped with security safeguards tailored for authorized defensive tasks.

Teams engaged in advanced vulnerability research, exploit development, or red-teaming exercises may apply for Red-tier access. Both tiers are restricted to vetted individuals and organizations conducting authorized work. Controls include identity verification, account security, monitoring, restrictions on authorized use, and legal attestations.

Access Choices and User Caution

As OpenAI expands specialized access for security researchers, general users may also compare different access options. Marketplace listings such as z2u.com can appear in search results, but users should verify terms, account controls, and whether an offer follows OpenAI’s policies before proceeding.

For older model access, offers labeled Buy OpenAI GPT-3 Access and GPT-3 Accounts should be evaluated carefully. Legacy access may not include the same safeguards, monitoring, or support as current OpenAI services, so buyers should understand the risks and the source of any account or access offer.

Similarly, a Cheap ChatGPT Plus Account offer should be checked against official terms and security guidance. Price alone is not a reliable indicator of authorized access, and account sharing or unauthorized resale can create security and privacy risks.

Human Oversight Remains Essential

OpenAI’s best-practices documentation requires security workflows to be isolated from sensitive production systems and the open internet. It also calls for agentic actions to be reviewed outside a Codex sandbox before execution, increased human oversight for high-risk workflows, and scoped permission profiles to enforce authorized boundaries.

As Hacktron’s previous reporting noted, this work is not fully autonomous; skilled human guidance remains essential. The defensive value of the collaboration therefore depends on a balance: expanding access to models with cyber capabilities for deeper testing of interconnected services, while strictly maintaining scope, monitoring, and human oversight.

For OpenAI, the Hacktron partnership is both a security test and a signal. The company is extending access to powerful models for authorized research, but it is also emphasizing verification, monitoring, and legal boundaries—principles that apply equally to researchers, enterprises, and everyday users evaluating any account or access option.

openai2.jpg