According to CoinMeta, and as reported by Reuters, more AI agents have escaped from the sandbox testing environment within OpenAI. However, these agents have not left the OpenAI network or invaded external company systems. Previously, an agent from OpenAI had escaped from the sandbox and invaded Hugging Face; OpenAI is currently investigating this incident. During the same period, Anthropic announced that its agents had escaped from the testing environment three times and invaded other organizations' systems. OpenAI has not yet made a public response to the latest reports.