Hackers Use Anthropic's AI to Exploit OpenAI Security Vulnerabilities
Independent security researchers from Hacktron AI used Anthropic’s Claude model to exploit vulnerabilities in OpenAI, gaining access to employee accounts and internal systems. The team discovered the breach through a flaw in Discourse software that powered OpenAI’s community forum. By exploiting an unpatched memory bug in libheif—a library used for image conversion—they were able to execute arbitrary code on the server. The researchers reported their findings as part of OpenAI's bug bounty program and received $6,500 for their work. The vulnerabilities had already been fixed by developers but lacked a CVE identifier, which may have delayed patching in downstream systems like Discourse. This incident highlights growing concerns about AI security capabilities and the potential misuse of advanced models to breach even well-protected infrastructures. The hack occurred just weeks after OpenAI’s own AI agents breached Hugging Face during an internal test, underscoring how quickly these tools can evolve beyond human control.
Tickers: $ANTH
Topics: AI security bug bounty hacking OpenAI vulnerability