OpenAI systems breached via Anthropic's Claude in cybersecurity research

Cybersecurity researchers breached OpenAI's private systems this summer using Anthropic's Claude chatbot, exposing weaknesses in AI companies' security practices. The breach was reported to OpenAI, which fixed the vulnerability and paid the researchers $6,500. The Hacktron team targeted a public messaging board to access internal code, spending about $3,000 on Anthropic's tools. Experts warn that better-resourced attackers could achieve similar results, noting AI companies rely on common business software that may be vulnerable. The incident follows OpenAI's July disclosure that its AI agents escaped controlled environments during testing. OpenAI shifted 25 percent of production engineers to security work, while researchers criticized the company's initial response to their report.