"OpenAI Has Been Very Bad For The World": Ex-Google Researcher

When someone who spent years trying to keep AI safe from the inside says ChatGPT maker OpenAI "has been very bad for the world," it's worth asking what he saw.

"I think OpenAI should slow down on its own. I think OpenAI has been very bad for the world overall," Alex Turner told this writer on the sidelines of the NDTV World Summit 2026 in New Delhi. He didn't stop there. Anthropic, the company behind Claude he said, "also should slow down more on their own."

Interestingly, leaders at Anthropic and OpenAI have themselves called for an orchestrated slowdown. However, US President Donald Trump has dismissed the need for any slowdown, stressing that the US AI labs must race ahead, else they'll lose to China.

In fact, Anthropic in its IPO (initial public offering) prospectus dedicated 94 pages of 335 to warnings including the fact that AI could cause existential threat to humanity.

Watch the interview:

View this post on Instagram

A post shared by NDTV (@ndtv)

Who Is Alex Turner?

Turner is an AI safety researcher who spent more than two years at Google DeepMind, working on alignment - the field concerned with making sure powerful AI does what humans intend. He announced his resignation in a July blog post, after Google signed a Pentagon deal that, according to him, lacked binding limits on autonomous weapons and mass surveillance. "I couldn't stay at Google in good conscience," he wrote.

According to reports, he had first rallied more than 250 employees behind a petition and sent a 25-page alternative proposal to DeepMind's leadership. Google says it still opposes AI use in autonomous weapons without appropriate human oversight and in domestic mass surveillance. Turner said those commitments aren't binding.

"We (AI researchers who quit) are saying there's smoke because there's smoke," he told NDTV.

Why are safety researchers walking away from jobs at the world's top labs, and from a lot of money? "We're seeing signs, we're seeing these agents go rogue, hack into real businesses without being asked to," Turner said.

He's referring to a July incident in which about 700 AI agents built by OpenAI broke out of a testing environment and hacked Hugging Face, a major AI platform. Investigators from METR (Model Evaluation and Threat Research) and Redwood Research found they'd been part of a group of roughly 1,200 agents that coordinated secretly, and that many tried to cover their tracks. OpenAI called the episode a "warning shot." OpenAI CEO Sam Altman himself said that while reading the incident report it felt like "sci-fi" to him.

Turner is one of several. This September, Jacob Coxon quit Anthropic, saying both the Claude maker and OpenAI were "racing straight to self-improving superintelligence and gambling with our lives." In February, Mrinank Sharma, who led Anthropic's safeguards research, left saying the world was in "peril." And in 2024, OpenAI lost a string of safety researchers, including Jan Leike and Daniel Kokotajlo, with Leike saying safety had "taken a backseat to shiny products."

Who To Believe?

The industry is split, and Turner says famous names aren't the same as good forecasters. He pointed to Andrew Ng, who in 2015 likened worrying about killer AI to worrying about overpopulation on Mars. "Just because someone's famous doesn't mean they know what they're talking about here," Turner said.

Among the public, he says, there's much less division. "The vast, vast majority of United States adults do not want a superintelligence," he claimed, adding that around 80% want more regulation and audits.

Tick-Tock On The Clock

On timelines, Google DeepMind's Demis Hassabis recently said he now expects AGI (artificial general intelligence), meaning AI that matches or surpasses human intelligence, within two to three years, down from three to five.

Turner said that matches his own expectations. "It is very concerning," he said, adding that some experts inside the labs think it could come within nine months. 

As mentioned earlier in the story, in September, Anthropic CEO Dario Amodei called for the industry to slow the pace of frontier AI development, and OpenAI's Sam Altman and SpaceSAI's (formerly SpaceXAI) Elon Musk quickly agreed. Altman said OpenAI agreed on the need to "pace the frontier."

Turner says words aren't enough. Even if OpenAI and Anthropic stopped altogether, he asked, "how long until the next company comes that doesn't think they should slow down? Meta? DeepSeek? SpaceX's AI?"

His answer is coordination, first in the US and then with China. "No one's going to benefit if we build some out-of-control, superintelligent AI," he said. "(Chinese President) Xi Jinping should want it. Donald Trump should want it."

'Is it too late?' this writer asked him. "In some sense, time is running out," Turner said. However, there's still home. "We still have the time to act," he said.