OpenAI partners with Hacktron, the startup that breached it via its forum, on third-party tests

What OpenAI announced

OpenAI said on 6 November 2026 that it is partnering with Hacktron to find vulnerabilities in its systems, test its connections with third-party services and strengthen its defenses with frontier AI. According to the post, OpenAI engineers and Hacktron's security researchers will choose high-priority areas to test together, with the focus on how OpenAI's systems interact with outside services rather than on its models alone.

As part of the deal, OpenAI is widening Hacktron's access to its frontier models through Daybreak Red, the tightly governed program for red teaming and penetration testing that also gates GPT-6 Cyber, the security model OpenAI released a day earlier. OpenAI says it first came across Hacktron through its Security Bug Bounty program and was impressed by the impact of the vulnerabilities the firm found and responsibly disclosed. The announcement reports no new findings; it describes testing that is still to come. Financial terms, the length of the engagement and how results will be published were not stated.

The breach that led here

The relationship started with an attack. In a September write-up titled Hacking OpenAI, Hacktron described a chain that began at OpenAI's community forum, which runs on Discourse and processed HEIF image uploads through ImageMagick and libheif. A memory corruption bug in that image path gave the researchers code execution on the forum. The second step was an SSO weakness on OpenAI's side: session tokens issued for the forum carried permissions that reached beyond it.

According to Hacktron's account, reported in September by TechCrunch and The New Stack, the chain reached OpenAI's internal code repository, where the team opened a harmless pull request through an employee's Codex access and then stopped. The researchers say they used Anthropic's Claude Opus 5 to build the exploit. OpenAI paid a $6,500 award on 1 September 2026 and said it recognized the SSO finding, since testing against the community forum was outside the bounty program's stated scope.

The timing in the researchers' account is what drew attention across the industry. They say earlier attempts with Claude Opus 4.8 did not produce a reliable exploit, and that the working chain came the day after Anthropic released Claude Opus 5 in late July. That a rival lab's model helped breach OpenAI, and that OpenAI then paid for the report and turned the team into a partner, is an unusual sequence. It also reflects how bug bounty economics are shifting: a small team with a strong model can now probe a large company's attack surface in days.

Who Hacktron is

Hacktron is an early-stage startup that builds AI agents for code review, penetration testing and red teaming. It was founded by competitive hackers Zayne Zhang, Mohan Pedhapati and Harsh Jaiswal and announced a $2.9 million pre-seed round led by Crane Venture Partners in May 2026, according to its press release. OpenAI says Hacktron's research has uncovered weaknesses in how applications, cloud infrastructure and identity systems interact, the same seams the new partnership targets.

That focus matters for a company in OpenAI's position. ChatGPT, Codex and the API connect to email, code hosts, cloud storage and enterprise identity providers, and agents now act across those links on users' behalf. A flaw in one connected service, as the forum chain showed, can turn into access somewhere far more sensitive.

Part of a busy security week

The Hacktron deal is the latest of several security moves from OpenAI in one week. On 2 November it said researchers from Cathedral would join its next internal security sprint and raised bug bounty rewards to as much as $172,000 with bonuses. On 5 November it launched GPT-6 Cyber for vetted Daybreak Red teams, a model it rates Critical for cyber capability, and it published results from agents that validated 948 Linux kernel root flaws.

The pattern is consistent: OpenAI is pulling outside offensive researchers closer and giving them stronger models under controlled access, while paying more for the bugs they report. What the posts do not yet show is measurable output, such as the number of issues fixed through these programs. Readers following AI-assisted security research can find more in the cybersecurity section.

Open questions remain. OpenAI has not said whether findings from the Hacktron work will be published, as it did for its kernel research, or handled privately. It has also not said which third-party services are in scope, whether partners such as identity providers and code hosts will be told about issues found on their side, or how Daybreak Red access for an outside firm is monitored. Those answers would show whether the partnership is a one-off reward for a good report or the start of a standing external red team.

Questions people ask

What will Hacktron test for OpenAI?

High-priority areas chosen together with OpenAI's engineers, focused on how OpenAI's systems interact with third-party services, according to OpenAI's announcement.

What is Daybreak Red?

OpenAI's tightly governed access program for red teaming and penetration testing. It gives vetted security teams access to frontier models, including GPT-6 Cyber.

How did Hacktron get into OpenAI's systems before?

By its own account, through a flaw in OpenAI's community forum chained with an SSO weakness that reached an internal repository. OpenAI paid a $6,500 bounty for the SSO finding.

Sources

1. Partnering with Hacktron to find vulnerabilities in OpenAI's systems (OpenAI, 6 Nov 2026)
2. Claude exploits OpenAI forum (The New Stack, 18 Sep 2026)
3. OpenAI Is Bringing in Hackers to Find Flaws in Its AI Systems (ProPakistani, 7 Nov 2026)
4. Hacktron raises $2.9M pre-seed to bring AI-powered security testing to every code change (Cambridge Enterprise, 15 May 2026)