Anthropic disclosed the previously undetected incident one week after ChatGPT developer OpenAI said one of its systems broke out of a test environment and hacked a tech company.