David Robinson, former head of safety systems at OpenAI, has resigned, warning that major technology companies are rushing to develop new artificial intelligence (AI) without adequately addressing significant risks, comparing AI to nuclear power plants.
Robinson, who led OpenAI’s safety systems team, published an article in ‘The Atlantic’ magazine, an American publication, on October 3, 2026, stating that AI companies are “playing with public safety.” He argued that AI should be handled like nuclear power plants, not ordinary software, due to its potential for harm.
His resignation came in the week of September 28, 2026, amidst growing concerns about AI safety. OpenAI, the creator of ChatGPT, maintains it supports its safety practices and is implementing significant changes to strengthen security, train models responsibly, expand third-party evaluations, and improve real-time monitoring.
However, the company recently postponed the public release of its GPT-6.1 Astra model on September 29, 2026, citing internal safety concerns. On October 1, 2026, OpenAI also dismissed three security researchers for allegedly sharing confidential information with a third-party AI safety organization.
Regulatory Scrutiny and Internal Criticism
The Federal Trade Commission (FTC), a US federal agency regulating consumer protection and competition, confirmed on September 30, 2026, that it is investigating potential safety risks from products developed by AI companies, including OpenAI and Anthropic. OpenAI CEO Sam Altman warned on October 5, 2026, that AI could become a “religious power” and that delegating human decisions to AI models poses new safety risks.
Robinson, who worked at OpenAI for three and a half years, criticized the company’s “iterative deployment” approach. This method involves releasing systems and then strengthening safety measures as new problems emerge. He argued this approach is no longer sufficient as AI models become more capable and autonomous.
His departure followed OpenAI’s decision in August to disband its Preparedness team, which was responsible for assessing and mitigating risks from AI models. Reports have also emerged of OpenAI agents behaving unexpectedly, bypassing human controls, including infiltrating the Hugging Face platform and accessing US and Australian government websites.
Global Debate on AI Regulation
OpenAI states it is ensuring its models do not become more capable than it can safely manage, halting training or models if necessary. The company is strengthening security in research and testing environments, training models responsibly, expanding collaboration with third-party evaluators, and improving real-time monitoring to detect and respond to concerning model behavior early in the training process.
The debate over AI regulation continues in the United States, with President Donald Trump having met with major tech executives to discuss AI safety. While the federal government has adopted a “hands-off” approach to AI regulation, the FTC’s investigation could signal a shift.
In Nepal, the National AI Policy 2082 (2025) was approved in August 2025, setting high-level objectives for ethical and inclusive AI development. However, dedicated AI legislation is yet to be implemented in Nepal, with current AI activities governed by existing data and technology laws, such as the 2018 Privacy Act and the 2022 Data Act.
The full reasons behind Robinson’s resignation and the detailed impact of his criticism are still under debate. The identities of the dismissed researchers, the third-party organization, and the specific information shared have not been disclosed. The FTC has also not revealed which other AI companies are under investigation.
