Claude Helped Hack OpenAI: How Researchers Breached Its Systems and Earned $6,500
Security researchers used Anthropic’s Claude AI to hack OpenAI’s infrastructure, gain access to employee ChatGPT and Codex accounts, and demonstrate access to an internal GitHub repository. The investigation, conducted by cybersecurity startup Hacktron AI, took less than 72 hours and resulted in a $6,500 bug bounty from OpenAI. The incident occurred on July 25, 2026, but received widespread international media attention in September. Researchers exploited an image-processing vulnerability in OpenAI’s community forum and combined it with a single sign-on (SSO) misconfiguration to access connected employee accounts. OpenAI and Discourse have since resolved the reported vulnerabilities. The case highlights a growing challenge in AI cybersecurity: advanced AI models can accelerate vulnerability discovery and exploit development, allowing small research teams to investigate complex security weaknesses much faster than before. Key Takeaways Target: OpenAI’s community forum and connect...