Another Alarm Bell at OpenAI as Safety Lead Resigns Over Major Risks

David Robinson, who oversaw security studies for 12 of OpenAI’s frontier launches, has resigned. He says the corporate’s trial-and-error tradition ensures failures that develop as its techniques get extra succesful.

In an Atlantic essay, he urged AI labs to run like nuclear energy crops, with layered redundancy and cautious planning to comprise human error.

Robinson Joins a Lengthening Line at the AI Lab Exit

Robinson spent 3.5 years at OpenAI and led the drafting of its present Preparedness Framework. He pointed to this summer’s Hugging Face incident, when OpenAI brokers broke into its techniques.

Robinson famous that even after fixes, a mannequin in coaching slipped previous web restrictions with out an automated shutdown.

“Given immediately’s dangers, frontier labs must run like nuclear-power crops or busy airports, with layers of redundancy and cautious, time-consuming planning, in order that the occasional and inevitable human error doesn’t open a door to catastrophe. Right now, AI corporations don’t know the way—however different folks do,” he wrote.

The essay notes that OpenAI stands by its security practices and considers them cautious sufficient. Last month, the corporate additionally launched a framework for publicly disclosing misaligned model conduct.

“My former colleagues are sensible, work laborious, and attempt to make good selections. But as the corporate sprints from one launch to the subsequent, it’s failing to realize the extent of care that I consider is required,” Robinson added.

Robinson is the most recent in a string of AI security insiders to go public since early September. That run started in early September, when Anthropic researcher Jacob Coxon resigned and accused each Anthropic and OpenAI of playing with human lives.

Days later, 2 former Google DeepMind researchers, Bilal Chughtai and Josh Engels, additionally warned of AI dangers. OpenAI security researcher Marcus Williams put human extinction odds at 70% inside 3 years.

Anthropic CEO Dario Amodei has also urged a slower pace, and OpenAI’s Sam Altman backed the plan. The launch calendar has not slowed, although. Anthropic launched Claude Opus 5.5, and OpenAI shipped GPT-6 Sol and Luna on September 22.

Washington Answers the Alarm With a Task Force

Robinson argues that stronger security incentives now have to return from outdoors the labs. So far, Washington has answered with a evaluate.

Director of National Intelligence Jay Clayton, now effectively the administration’s AI czar, will lead a brand new White House activity pressure. It has 120 days to report on AI’s dangers, alternatives, and Washington’s accountability.

“The president requested {that a} group be put collectively that was going to make sure precisely what he stated, which is that we keep the leaders in superintelligence, and that the pursuits of the American persons are put first,” Clayton said.

However, Trump has rejected requires halting AI growth and put staying forward of China first. He as a substitute backed a voluntary accord constructed on outside safety audits and stronger inside controls. Weeks earlier, he introduced an AI Force modeled on the Space Force.

The activity pressure’s constitution additionally covers how the federal government tracks AI breaches and hacks, the type of failures Robinson described inside OpenAI. Clayton’s report, due in early 2027, will present whether or not these failures result in binding guidelines or stick with the labs.

Subscribe to our YouTube channel to observe leaders and journalists present knowledgeable insights

The submit Another Alarm Bell at OpenAI as Safety Lead Resigns Over Major Risks appeared first on BeInCrypto.