Researchers Hack OpenAI With Claude

Independent security researchers from Hacktron AI used Anthropic’s Claude chatbot to break into OpenAI employee ChatGPT accounts and reach OpenAI’s internal software environment. The entry path ran through Discourse, the third-party software behind OpenAI’s community forum, then through weaknesses in OpenAI employee validation and single sign-on. The researchers said the work was an authorized bug hunt under OpenAI’s reward program, and they reported the flaws instead of abusing them. OpenAI said it thanked them, narrowed community sign-in permissions, revoked affected tokens and sessions, and fixed the vulnerabilities. Hacktron said the chain took less than 72 hours. OpenAI paid a $6,500 bounty. The case now feeds wider concern about cyberattacks on leading AI firms.