OpenAI has released a model built to find security vulnerabilities, deliberately loosened the restrictions that would normally stop it writing exploit code, and restricted who can use it. The results ...