Ars Technica · AI· 22:22OpenAI says planned GPT-6.1 is too insecure to releaseSimilar performance, security trade-offs also seen in current public models.#媒体#AI#alignment