Third-party cyber evaluations involving OpenAI models36gglub about 4 hours ago 4 commentsRead Article on openai.com
Discussion (4 Comments)Read Original on HackerNews
Smoke out those misconfigurations while the model only needs to escape, not do anything once out.