OpenAI Scrambles to Fix Safety Issues Before Shipping Next Flagship Model

OpenAI is working to make its next flagship model safe to release, as its use of loop transformers (recurrent depth) makes it harder to monitor how the AI reasons. Microsoft accidentally confirmed the technique was used in GPT-6.1 Sol before editing its blog post.

• OpenAI uses loop transformers (recurrent depth) in its next model
• The technique makes AI reasoning harder to monitor
• Microsoft accidentally confirmed its use in GPT-6.1 Sol
• The company edited the blog post to remove the mention