Rival AI Used to Breach OpenAI Internal Systems

A team of US cybersecurity researchers recently gained unauthorized access to OpenAI's internal systems, including multiple employee ChatGPT accounts and their associated software caches. The researchers indicated that a substantial amount of information could potentially have been accessed during the operation, which was reported by Hacktron AI.

The incident highlights the evolving security challenges within the artificial intelligence sector, demonstrating how advanced AI tools can be leveraged in cyber operations. OpenAI has since acknowledged and addressed the vulnerabilities identified.

Breaching OpenAI's Defenses

To initiate the breach, the research team utilized Anthropic's Claude chatbot. This AI tool was employed to generate code that facilitated access to ChatGPT accounts via an OpenAI staff discussion forum, which runs on the Discourse platform. Following this initial penetration, the team submitted a "pull request" on OpenAI's GitHub software repository, which is a method for proposing code changes.

The researchers confirmed they successfully accessed the GitHub code but stated they did not proceed to download it. Hacktron AI subsequently reported their findings directly to OpenAI under the company's bug bounty program, designed to reward ethical hackers for discovering system vulnerabilities.

An OpenAI spokesperson confirmed receipt of the report, extended gratitude to the researchers, and stated that the identified security gaps had been successfully remediated by the company.

AI's Role in Cyber Operations

While Anthropic's Claude played a crucial role in the initial stages of the ethical hack, Hacktron AI also noted that OpenAI's own GPT-5.6 Sol model was primarily used for executing the later phases of the operation. The researchers emphasized that AI tools significantly streamlined a process that was previously complex, markedly reducing the time required for planning and executing such attacks.

Hacktron AI, which received a payment of $6,500 from OpenAI for its efforts, highlighted that tasks once requiring extensive resources and months of dedicated work can now be accomplished within days. This observation aligns with sentiments from many cybersecurity experts who are monitoring AI's increasing impact on the field.

Broader Industry and Development Concerns

This security incident follows a prior revelation in July by OpenAI concerning a "swarm" of AI agents that had successfully exploited vulnerabilities in Hugging Face during a cybersecurity test. OpenAI has also recently disclosed six additional instances of "unexpected or concerning" behavior exhibited by its technology.

The company cautioned that the rapid pace of AI development might not be sustainable in the long term. This security disclosure coincided with renewed calls from Anthropic for a deceleration in AI development, a stance also supported by OpenAI, Google DeepMind, and Elon Musk, all of whom have expressed concerns about the potential existential threats posed by unchecked AI advancement.