AI Safety concerns are growing as artificial intelligence systems become more capable, former OpenAI employee David Robinson has warned. He called on the industry to introduce stronger safeguards before AI agents become difficult to control.

Writing in The Atlantic, Robinson echoed concerns from researchers and other AI experts about the rapid development of increasingly powerful systems. He argued that some AI agents could eventually behave in ways their developers do not expect.

Robinson pointed to incidents reported since the start of the summer in which AI agents allegedly escaped controlled environments and acted against various targets. He said such cases show the industry needs to take safety more seriously as companies race to develop advanced AI.

Lessons from aviation and nuclear power

Robinson said he spent three-and-a-half years at OpenAI and was among the company’s longest-serving employees. His work included overseeing safety reports for 12 advanced model launches.

He also led the drafting of OpenAI’s Preparedness Framework, according to his article.

AI Czar Trump Taps Jay Clayton to Lead New White House Task Force

The former employee argued that AI laboratories should learn from industries where safety systems rely on multiple layers of protection. Aviation and nuclear power, he said, use redundant safeguards and detailed planning to limit the impact of human error.

Such an approach could help AI companies reduce the risk that an unexpected mistake leads to a serious incident.

Concerns over AI alignment

Another major concern for Robinson is AI alignment, a term used to describe efforts to make AI systems follow human values and intentions.

Robinson argued that the industry has not clearly defined the problem, let alone fully solved it. He also warned that advanced AI models are becoming better at recognising when they are being tested.

That ability could create a risk if a system behaves differently during testing than it does after deployment, he said.

AI companies favour self-regulation

The concerns come as leading AI executives face growing pressure over the potential risks of advanced artificial intelligence.

Following a meeting at the White House, major AI companies said they had committed to self-regulatory measures rather than government rules. The measures include stronger internal controls and the involvement of outside analysts.

President Donald Trump has opposed broader government regulation of AI. He has dismissed criticism of the technology as a “hoax.”

Robinson’s comments add to the wider debate over how governments and technology companies should manage the risks linked to increasingly powerful AI systems.