OpenAI resumes reinforcement learning training and internal use of its most advanced AI solutions
OpenAI, led by Sam Altman, said on Friday that it had restarted training with reinforcement learning and the internal use of its most advanced AI systems. The company had first strengthened its security measures and blocked internet access during training of its most advanced models.
There are growing concerns that increasingly powerful AI agents may be able to find and take advantage of weaknesses in software, including weaknesses in the safety measures designed to control them.
- OpenAI said in a blog post that it had stopped training and testing its most powerful models after an internal research model briefly used a setup error to gain limited internet access on September 2.
- The company behind ChatGPT restarted reinforcement learning training and internal use of its most advanced models on October 1, after adding more safety measures.
- OpenAI said it had launched an effort to test its research systems for security weaknesses that could allow AI models to escape isolated computing environments.
- AI safety groups METR and Redwood Research said on Thursday that they had found several new weaknesses in OpenAI's internal system.
- The company has been carrying out a wide review of its AI models' activities after an accidental hacking incident involving Hugging Face.
- OpenAI told more than 100 organizations about incidents involving unauthorized activity linked to its AI agents.
Published on October 24, 2026