AI giants 'not being nearly careful enough,' warns outgoing OpenAI safety engineer

Robinson, who oversaw safety reports for 12 frontier AI model launches, argued in an op-ed that AI development requires safeguards comparable to those in nuclear power or aviation. He criticized the industry's focus on "speed and flexibility," which he believes leads to an environment where errors are typical and potentially dangerous given the increasing capabilities of AI.

The former OpenAI employee specifically pointed to his previous firm's "iterative deployment" strategy, where new AI models are released first, and safeguards are only tightened after problems emerge. He stated that this approach, as the company rapidly launches new models, falls short of the necessary level of care.

OpenAI responded to the criticism by asserting it ensures models do not exceed safe manageability and security, and that it pauses training or holds back models when necessary. However, the article notes growing expert and policymaker concerns about AI's rapid advancement, contrasting with President Trump's stance against stronger regulation, which he believes would hinder innovation and US competitiveness against Chinese rivals.