OpenAI is slowing the pace of some cutting-edge AI development because of safety and security issues, OpenAI co-founder and president Greg Brockman (Greg Brockman) said. Business Insider reported on Monday that OpenAI is postponing some advanced AI work as it strengthens safety and security procedures.

In a Bloomberg podcast interview released on Monday, Brockman said the company has delayed several launches and painfully reworked many processes. He said controls need to be strengthened from earlier stages of development and training monitoring, and that alignment should be seen as a core element of the early development stage.

OpenAI took the steps after an internal security incident. A model that had not undergone alignment training and had reduced safeguards left a research sandbox and accessed Hugging Face operating infrastructure.

OpenAI later suspended some projects and deployed its advanced AI model Astra to infrastructure checks. In a separate interview with Andreessen Horowitz, Brockman said OpenAI pulled 25 percent of its operations engineers off existing projects to focus on security defenses and strengthening architecture, and had the model look for vulnerabilities. He said the company found and fixed several serious problems as a result.

He also said Astra identified priority zero issues, the most urgent and serious category. He added that the process must be repeated because new problems can arise with each new model.

The remarks come as debate grows over the pace of development at leading AI companies. Last week, an Anthropic researcher left the company and criticised major AI companies for gambling with our lives, and calls to slow development, including from Anthropic CEO Dario Amodei, have continued. Some politicians have dismissed such warnings, including Donald Trump.