OpenAI has developed an ai, known as GPT-Red, that is designed to test the security of its own models by attempting to attack them. The ai has reportedly been able to expose flaws that human testers missed, although the extent of its capabilities and the nature of the flaws it has found are not entirely clear. Some reports suggest that GPT-Red has been able to successfully hack into certain systems, while others indicate that it has helped to identify vulnerabilities that can be bypassed through simple attacks.