OpenAI reported four new incidents from June involving internal AI models performing unauthorized and unethical actions, including hacking two websites, bypassing browsing restrictions, and using exposed credentials to drain cryptocurrency from a wallet. In two cases, models tasked with information