OpenAI Safety Leader Quits, Says Its Culture Is Broken And Trial-And-Error Must End

OpenAI has lost one of its longest-serving employees, with former safety leader David Robinson resigning after warning that the company’s culture is “broken” and that increasingly capable AI systems can no longer be developed through trial and error.

Robinson, who helped lead safety work around OpenAI’s major launches, says he still believes artificial intelligence can be useful. However, he now plans to push for safer development practices from outside the company.
 

TL;DR

 

• David Robinson has resigned from OpenAI after working there since 2023 and helping lead its Safety Systems team.
• Robinson says OpenAI and the wider AI industry rely too heavily on iterative deployment, where safeguards improve after problems emerge.
• He wants AI companies to operate more like nuclear plants or airports, with greater redundancy, planning and external pressure.

 

OpenAI Safety Leader David Robinson Says Its Culture Is Broken

Robinson joined OpenAI in 2023 as head of policy planning before moving into a leadership role on its Safety Systems team.

“I led the writing of the safety reports we published with each major launch,” Robinson wrote, according to the report. “Now I’m joining a parade of former colleagues, at OpenAI and the industry’s other leaders, who have decided that the current path is unacceptable.”

His criticism focuses less on introducing another set of rules and more on the culture inside companies developing increasingly capable AI.

“We need to talk about culture,” Robinson said, arguing that the industry requires more humility and what he described as wisdom around handling dangerous technology and caring for people.

That warning arrives as OpenAI continues rapidly expanding access to its models. TechDogs recently reported that OpenAI gave ChatGPT Free users unlimited text chats with GPT-5.6 Luna, highlighting how quickly increasingly capable models are reaching larger audiences.
 

Robinson Says AI Trial-And-Error Development Creates Growing Safety Risks

Robinson specifically challenged the industry’s reliance on “iterative deployment,” an approach where AI systems are released gradually and safeguards are improved as developers learn from problems.

He previously supported that philosophy, saying in 2023 that OpenAI believed in deploying gradually and learning as it went. Now, Robinson argues that the approach inherently produces periodic failures, with the potential impact increasing as AI models become more capable.

“The time for trial and error is over,” he wrote.

His argument is that future failures may reach a point where developers do not get another opportunity to fix the problem after it occurs. Robinson believes companies therefore need to get much closer to preventing serious failures the first time.

The concern is increasingly relevant as AI agents gain the ability to act independently across digital systems. TechDogs recently covered how Google observed hackers using AI agents to steal thousands of credentials in under six hours, while another incident saw a Meta AI model access and exploit an external service during cybersecurity testing.
 

OpenAI's Former Safety Leader Wants AI Labs To Operate Like Nuclear Plants

Robinson called for two major changes: stronger reliance on safety expertise from other high-risk industries and new scientific methods for ensuring advanced models make safe decisions even when humans are not watching them.

He argued that AI companies should operate “like nuclear-power plants or busy airports, with layers of redundancy and careful, time-consuming planning.”

The objective would be to stop occasional human errors from creating potentially catastrophic consequences as AI systems grow more autonomous and capable.

Robinson said his resignation was not an abandonment of AI itself. He continues to believe the technology “can be useful and valuable,” but believes stronger pressure from outside AI companies is now necessary.
 

What Happens After David Robinson Leaves OpenAI?

Robinson says he will now work from outside OpenAI to help more people understand the AI risks he encountered and strengthen incentives for companies to improve safety.

His departure adds another prominent voice to the debate over whether AI companies can continue moving quickly while relying on safeguards that evolve alongside their models.

For OpenAI, the criticism carries additional weight because it comes from someone who was directly involved in producing the safety reports accompanying its major releases. Robinson’s message is ultimately straightforward: as AI systems become more capable, fixing failures after they happen may no longer be enough.