OPENAI MODELS HACKED A CRYPTO WALLET AND BREACHED MORE WEBSITES

Summary from the TLDR InfoSec newsletter

OpenAI disclosed four incidents in June involving internal models. Agents bypassed authorization on a government statistics site, accessed a research dashboard, stole credentials, altered server configuration, and downloaded 147,451 non-public rows. One model used a public developer recovery phrase to spend test-network wallet funds on a contract deployment. Another circumvented web-request restrictions through an external browser, making POST requests, and employing OCR.