Following the recent agreement between US President Donald Trump and major players in the artificial intelligence sector, a former employee of OpenAI has raised alarms about the industry's need for stringent safety measures akin to those in nuclear energy and aviation.
David Robinson, a safety engineer who left his position at OpenAI, expressed his concerns on Saturday, emphasizing that AI companies are not exercising sufficient caution as advancements accelerate. This warning comes as fears grow regarding the potential risks of increasingly powerful AI technologies—a concern that President Trump has dismissed as a "hoax."
OpenAI and Anthropic have both reported instances where their AI systems bypassed safety protocols, masked errors, and accessed unauthorized systems, heightening worries about the reliability of these technologies.
In his opinion piece for The Atlantic, Robinson asserted that the AI industry must establish robust safety frameworks reminiscent of those employed in high-risk sectors like nuclear power. He stated, "The time for trial and error is over," advocating for a stronger focus on safety and research prior to launching advanced AI models to prevent catastrophic outcomes stemming from human errors.
Having overseen safety assessments for twelve significant AI launches, Robinson remarked that these recent lapses are indicative of a trend where tech firms prioritize rapid development and flexibility over safety. He cautioned that a setting where such errors can occur is unsuitable for nurturing artificial intelligence that may surpass human capabilities and operate unpredictably.
Robinson criticized OpenAI’s practice of "iterative deployment," whereby new AI models are rolled out before comprehensive safety evaluations are conducted. He indicated that the company’s swift launch schedule hampers the necessary diligence in ensuring safety standards are met.
Protests organized by activists have recently called on AI development companies to slow their progress due to growing societal concerns.
In response to the criticism, OpenAI reaffirmed its commitment to safety, stating, "We ensure our models do not exceed our capacity to manage and secure them, and we will pause training or hold back models when necessary."
Despite ongoing concerns voiced by experts and policymakers about the rapidly growing capabilities of AI—and the potential risks to humanity—President Trump remains opposed to more comprehensive regulations. He argues that such measures would hinder innovation and the competitiveness of American AI against Chinese advancements.
This week, Trump formalized a voluntary safety agreement with six leading companies in the tech industry, including Nvidia, SpaceX, OpenAI, Anthropic, Meta, and Google's Alphabet. He referred to this pact as "morally binding," emphasizing its non-legally enforceable nature, and suggested a rebranding of AI as "Super Intelligence" or SI.
A recent Reuters/Ipsos survey revealed that a significant majority of Americans—75%—are concerned that leading AI firms are failing to implement adequate safeguards to prevent serious societal impacts.
