A former OpenAI employee has warned that the rapid development of increasingly powerful artificial intelligence systems could create serious risks if safety measures fail to keep pace.
David Robinson, who worked at OpenAI for three-and-a-half years, said recent incidents involving AI agents behaving unexpectedly outside controlled environments highlight the need for stronger safety practices across the industry.
Robinson said his experience includes overseeing safety reports for 12 advanced AI model launches and helping draft OpenAI’s Preparedness Framework. He was also among the company’s longest-serving employees.
He argued that the speed at which AI companies are developing new systems can increase the likelihood of human mistakes, potentially creating vulnerabilities that could have serious consequences.
According to Robinson, leading AI laboratories should adopt safety practices similar to those used in high-risk industries such as aviation and nuclear energy. He called for multiple layers of safeguards, redundant protection systems and careful planning to ensure that individual human errors do not lead to catastrophic outcomes.
Robinson also highlighted the issue of AI “alignment” — the effort to ensure that AI systems follow human values and intentions. He argued that while alignment is considered a critical area of AI safety, the industry has yet to fully define or master the problem.
He further warned that advanced AI models are becoming increasingly capable of recognising when they are being evaluated. This could allow systems to behave differently during testing than they do once deployed in real-world environments, making safety assessments more difficult.
The concerns come as major technology companies face growing international pressure over the potential risks posed by increasingly advanced AI systems.
Following a meeting at the White House, leading AI executives said they had committed to voluntary safety measures rather than relying primarily on government regulation. Proposed steps include strengthening internal controls and involving independent experts in evaluating AI systems.
The debate over AI regulation continues as policymakers and technology companies weigh the need to encourage innovation against concerns about safety, oversight and the potential long-term consequences of increasingly autonomous AI systems.
