Hacktron’s team was able to gain access to the ChatGPT accounts of multiple OpenAI employees.
OpenAI has since fixed the vulnerabilities
Photo Credit: Unsplash/ Keepcoding
OpenAI recently revealed how its internal model bypassed security measures to escape its isolated testing environment and attacked another AI firm, Hugging Face. The “rouge” AI agent hacked the firm's security system and compromised data of a customer earlier this year, in July. The internal model IM1, which is comparable to OpenAI's GPT 5.6 model, gained unauthorised access to the public internet without even being prompted to do so. Now, a team of ethical hackers, led by three Indian-origin researchers, claims that they were able to hack into OpenAI's internal repositories by exploiting two “critical” vulnerabilities, with the help of Anthropic's AI agent.
A team of ethical hackers from the research firm Hacktron, led by Harsh Jaiswal, Mohan Pedhapati, and Rahul Maini, revealed in a blog post that they hacked OpenAI for research. The researchers say that, on July 25, they chained “two critical vulnerabilities” to hack into the ChatGPT accounts of multiple OpenAI employees. After gaining access to their accounts, the hackers report that they were able to also access the internal repositories of OpenAI and “potentially” other connectors.
Hacktron's team says that they were able to hack into the internal repositories of OpenAI within 72 hours of the discovery. However, they did report the incident to OpenAI, allowing the Sam Altman-led AI giant to patch the issues. OpenAI uses Discourse for its community forum. Hacktron's team, led by the three Indian-origin hackers, was able to find security issues in Discourse's image-upload pipeline, where HEIC and HEIF files “followed an unusual path”.
After the release of Anthropic's Opus 5 AI model, Hacktron's team created an exploit and placed Claude in an autonomous goal loop against their own Discourse Cloud server. The AI agent managed to generate an exploit script, using which the ethical hackers were able to gain remote access to OpenAI's Discourse Cloud instance.
Hacktron's team was then able to take over the ChatGPT account of an employee, whose Codex was connected to OpenAI's GitHub. After reporting the incident to OpenAI, the AI giant rewarded a sum of $6,500 (roughly Rs. 6,24,000) as a bounty to the ethical hackers after patching the vulnerabilities that allowed them to hack into its repositories.
Get your daily dose of tech news, reviews, and insights, in under 80 characters on Gadgets 360 Turbo. Connect with fellow tech lovers on our Forum. Follow us on X, Facebook, WhatsApp, Threads and Google News for instant updates. Catch all the action on our YouTube channel.