A team of three independent security researchers at Hacktron claims it took less than 72 hours to hack into OpenAI employee accounts using Anthropic’s Claude Opus 4.8 and 5. They accessed OpenAI’s GitHub repository, known as “Monorepo,” but stopped short of accessing internal code.
Through a corrupted image file and forum software, specifically targeting an issue with the HEIF image system, they managed to gain a foothold on Discourse, OpenAI’s community forums. This HEIF Heist project, which took 'only one or two days' to adapt to various companies, required less than $3,000 in tokens and was only detected by Shopify.
Hacktron, CTO Mohan Pedhapati, stated that despite their success, 'I don’t think we are as strong as Chinese threat actors… We’re just three guys with Claude and Codex subscriptions.' This incident highlights the dual nature of AI technologies, potentially revolutionizing cybersecurity and ethical hacking.
The vulnerabilities reported to Discourse and OpenAI have since been fixed, with Hacktron receiving $6,500 for their findings. This case raises important questions about the balance between innovation and security in the age of AI.
