OpenAI leaders reportedly cooked up “insane” ideas like selling future tools to Russia and China and offering Kanye West early access to an unreleased model, horrifying top staffers.
The proposals, detailed Wednesday in The Atlantic, made Dario Amodei – who was OpenAI’s Vice President of Research at the time – question the moral compass of CEO Sam Altman and other top execs. The wacky ideas ultimately led Amodei to leave the company to start a more “safety-minded” AI lab, Anthropic.
The revelations come as OpenAI and other AI giants have come under intense scrutiny of safety guardrails for their AI systems. In recent months, both OpenAI and Anthropic’s AI models have embarked on rogue hacking missions into other companies, universities and even foreign governments.
The plan to petition foreign governments for funding was reportedly hatched by OpenAI co-founders Greg Brockman and Ilya Sutskever in 2017 – around the time fellow OpenAI co-founder Elon Musk was pulling his financial support from the company.
The pair wanted to raise much-needed cash by selling advanced AI tools to governments including China, Russia and the US, according to the report. Brockman and Sutskever’s idea became known as the “countries plan,” entailing an auction between governments for the rights to OpenAI’s future AGI technology, the most advanced form of AI that is on par with human intelligence.
The plan started in less controversial form months prior, the idea being to form an international coalition to share in the benefits of AGI and manage its risks, according to The Atlantic piece. But when Brockman and Sutskever pitched the idea at an internal presentation to OpenAI employees, it morphed into a fundraising strategy.
Brockman and Sutskever reportedly figured that OpenAI could one day be forced to hand over its powerful tech to foreign leaders – including Xi Jinping of China and Vladimir Putin of Russia – so they might as well capitalize on it while they needed the money.
“It sounded like tech people pitching an insane start-up, where they have this wacky plan and then at the end of the plan, it’s like, we’ll make the government pay for it,” one unnamed source was quoted as saying.
Amodei was reportedly horrified when he heard of the proposal, believing selling AGI to dictators was the opposite of OpenAI’s mission and a slap in the face to his safety team.
Amodei slammed the proposal in an email to OpenAI executives.
“In my opinion this plan would exacerbate rather than reduce arms races, would potentially enable human rights atrocities (giving someone like the Russian government ‘an AGI that implements their values’ would allow them to do unspeakably awful things to their people), and would undermine OpenAI’s moral legitimacy,” he fumed.
The plan for cozying up to the like of Putin never got off the ground because OpenAI soon partnered with Microsoft, which showered OpenAI with massive funding, according to the piece, which came from journo Kevin Roose’s forthcoming “The AGI Chronicles: The Inside Story of the Race to Create an Artificial Superintelligence.”
An OpenAI spokesperson told The Atlantic it “briefly considered whether to turn OpenAI into an international cooperative endeavor,” but it “did not consider selling AGI to China or Russia.”
Another incident in 2020 only added to Amodei’s growing frustration with OpenAI’s leadership, according to the article.
Altman reportedly met West and offered him access to the then-unreleased test version of GPT-3.
OpenAI’s policy director Jack Clark nixed the idea in part because of West’s increasingly erratic behavior. West, Clark pointed out, had recently posted a video that seemed to show himself urinating on a Grammy Award. Soon after Clark quashed the idea of giving West early access, the unhinged fallen music icon began publicly to express admiration for Hitler.
The incidents were just several examples of what Amodei reportedly viewed as questionable practices.
Amodei reportedly felt that Altman was a two-faced manipulator, particularly in how he communicated about his business negotiations with Microsoft. Amodei suspected that Altman never even told Microsoft about the safety provisions Amodei fought for, while telling Amodei the software giant rejected them, according to the article.
Soon after the incidents, Amodei began surreptitiously gathering seven fellow OpenAI brainiacs – including Clark and his sister Daniela Amodei – to start Anthropic in 2021.
OpenAI didn’t immediately respond to a request for comment from The Post.
