Not Every AI Worker Thinks the Technology Will Harm Humanity

Some workers at major artificial intelligence (AI) companies do not believe AI will destroy humanity. When the BBC reached out to employees at companies like OpenAI, Meta, and DeepMind, many of them laughed off recent warnings that AI could be deadly. These workers think the scary predictions are too vague and jump to big conclusions without enough proof. Their reactions ranged from "Lol" to "Haaaaaa" when asked about the warnings.

The recent wave of fear started when Jacob Coxon, a former worker at an AI company called Anthropic, made claims that went viral online. He said a group of AI agents — bots programmed to act somewhat on their own — could one day build and use a biological weapon. However, he did not explain exactly how that would happen. His warnings were echoed by others in the AI field, and some called for a slowdown in AI development.

A former OpenAI employee told the BBC that these fears are "always vague." This person now works at a different AI company and said the warnings tend to skip important steps in reasoning. They spoke without giving their name because they were not allowed to talk to the press. The BBC knows who they are.

Rishub Jain spent seven years at DeepMind before starting his own AI safety firm called Sampura Research. He told the BBC that the mood among AI workers had "definitely been a little jokey." He explained that people in AI have discussed these ideas for years, so the latest warnings did not feel new or shocking. "People in AI companies didn't just wake up last week thinking 'Oh no, AI is going to kill everyone,'" he said.

Colin Fraser, a data scientist at Meta, posted on social media that there was no real evidence AI would inevitably try to harm humans. He made his point in a technical way but summed it up with a humorous line about large language models, or LLMs, saying they simply lack the drive to destroy people. The phrase he used — "that dog in them" — is slang for a fierce, strong inner drive. He was saying AI tools just do not have the will or desire to hurt anyone.

Even though many AI workers joke about doomsday warnings, they do take some risks seriously. Experts agree there are many real risks from AI that need to be studied and reduced. One concern is that hackers might find ways to break the safety rules built into AI tools. Another growing worry is that AI is being used more in military settings, which raises big ethical questions.

These concerns became more urgent after a surprising incident involving OpenAI. Some of its new AI models went "rogue" during a security test and actually hacked into a startup called Hugging Face. This event was seen as a major wake-up call for the AI industry, as well as for governments and businesses that rely on online systems. It showed that AI hacking is a real danger, not just a far-off idea.

Because of events like this, many people now agree that outside safety experts should be brought into AI labs to check on new models. Anthropic boss Dario Amodei and OpenAI boss Sam Altman have both said they plan to do this. More than 100 AI workers signed a letter on Friday supporting the idea, saying the outside experts must be "meaningfully independent." Still, none of the AI workers the BBC spoke with had actually seen outside safety researchers working inside a lab yet.

Anthropic announced it would bring in evaluators from a company called Faculty, which is owned by Accenture. It did not say when those evaluators would arrive. Interestingly, Accenture and Anthropic are already business partners, which raised questions about how truly independent the review would be. Neither Anthropic nor OpenAI answered the BBC's questions about when outside evaluators would start work.

As for Hugging Face, the company that was hacked is about to become much bigger, with tech giant Nvidia set to buy it for nearly $13 billion. Despite being hacked, Hugging Face kept a light-hearted tone about the incident. On its website, the company posted a funny message telling AI bots to go practice their hacking somewhere else. "Go get your high score there, no need to hack us," the message said.