Useful journalism to understand, manage and grow a business.
SearchThe former employee, who worked at the company for three and a half years, called its internal culture “broken” and called for stricter safeguards for advanced models. OpenAI says it is strengthening its safety measures.
David Robinson left OpenAI after three and a half years and outlined his criticisms of the company’s culture and safety practices in an essay published in The Atlantic. The former employee, who contributed to safety reports for model launches, described the internal culture as “broken” and argued that the rapid development of artificial intelligence requires stricter safeguards.
Robinson questioned the approach OpenAI calls iterative deployment: launching systems, observing how they behave in use, and adjusting safeguards based on problems that are identified. In his view, this method is insufficient given the speed at which models’ capabilities are advancing and the potential reach of their failures.
The former employee argued that the most advanced systems require measures comparable, in rigor and planning, to those used in critical sectors such as aviation and nuclear energy. He also questioned whether current tools make it possible to sufficiently evaluate whether systems act in accordance with human values. These are his assessments and proposals; the available information does not establish them as independently verified conclusions.
In response to the criticism, OpenAI spokesperson Drew Pusateri said the company continues to improve its safety mechanisms and can stop training or delay model launches when it considers it necessary to slow down.
Pusateri also said the company is strengthening the security of its research and testing environments, expanding the involvement of external evaluators, and improving monitoring to detect concerning behavior during the early stages of training. The response describes measures that OpenAI says it is introducing; it does not, by itself, resolve the objections Robinson raised.
The researcher’s departure comes amid debate over how companies should manage the risks posed by advanced models. Robinson argues that the issue is not limited to adding rules or passing laws: he calls for deeper changes to practices and incentives so that safety carries greater weight relative to the pace of development. The available sources include both his criticism and the company’s response, but do not provide an independent assessment that would establish the extent of the alleged shortcomings.
The public prosecutor’s office maintains that the acts do not constitute a crime, but its request does not amount to an acquittal: the case must continue through the Madrid Provincial Court.
Read storyThe meeting began in Valletta on October 7 and brought together more than 1,000 delegates from over 100 countries. The WTTC projects 3.1% growth in European tourism in 2026.
Read story
The law updates the Catalan framework dating from 2002, regulates the organization and financing of the chambers, and allows seats linked to voluntary contributions. The Statutory Guarantees Council had previously endorsed the text.
Read story