While the artificial intelligence industry is growing rapidly, security concerns are also increasing at the same rate. Internal discussions in leading companies such as OpenAI have brought to the agenda again the question of how much risk the development pace that pushes the boundaries of technology poses for humanity.
An experienced name who left the company's security staff announced that he resigned, saying "the culture was broken" because of this pace. David Robinson's open letter published in The Atlantic is not only a personal decision, but also the entire industry's question "speed or security?" It is a strong warning that opens the dilemma to discussion again.
David Robinson worked on security teams at OpenAI for three and a half years. During this period, he was among the names that prepared the company's "preparedness framework" and supervised the security reports of 12 frontier model launches. However, he argued that the company's recent rush from one launch to another prevented it from reaching the required level of care.
In his article titled “I Left OpenAI Because Its Culture Is Broken” published in The Atlantic, Robinson said: “As the company rushes from one launch to the next, in my opinion, it is not achieving the level of care needed.”
“THE ERA OF TRIAL AND ERROR IS OVER” WARNING
Robinson's harshest criticism is directed at OpenAI's "iterative deployment" strategy. In this approach, systems are released to the market and security measures are strengthened as problems arise. The former employee argued that advanced AI systems must be developed with the same stringent safety standards now applied to nuclear power plants or the aviation industry. “The era of trial and error is over,” said Robinson, emphasizing that artificial intelligence companies (including OpenAI) have not been careful enough and should prioritize security expertise and research before developing more capable systems.
TALENT PROGRESS FASTER THAN ALIGNMENT RESEARCH
Robinson also noted that artificial intelligence capabilities are advancing faster than the scientific understanding of “alignment,” which seeks to ensure that systems behave in harmony with human goals and values. This non-compliance, combined with incidents where security controls fail or experimental systems exhibit unexpected behavior, has increased industry concerns, he said. It was reminded that OpenAI and its rival Anthropic were also criticized for similar security vulnerabilities.
“WE STOP EDUCATION WHEN NECESSARY”
An OpenAI spokesperson defended the company's approach. "We ensure that our models do not become more capable than we can safely manage and secure. When we need to slow down, we stop training or withdraw models," the statement said. While this statement shows that the company continues to claim controlled progress, the former employee's criticism reignited security debates in the industry.
David Robinson's resignation and article once again revealed how fragile the balance artificial intelligence companies establish between speed and security is. While the industry produces more powerful models, it also has to establish more robust security architectures. Otherwise, the price of the “trial and error” period may be not only the reputation of the companies but also the social acceptance of the technology.
