AI Employees Push Back on Warnings That Artificial Intelligence Could Doom Humanity

Not all employees at major artificial intelligence companies believe unchecked AI development will lead to mass casualties or the end of humanity.

In text exchanges and conversations with the BBC, several people who have worked at companies including OpenAI, Meta and DeepMind expressed skepticism about claims that increasingly powerful AI tools could eventually kill people on a massive scale.

“LOL,” “Aaah” and “Bring Lulu” were among the reactions the BBC received after recent high-profile warnings from some AI industry insiders.

Former AI employees question existential-risk warnings

Concerns about the potential dangers of advanced AI date back decades. However, claims made last week by former Anthropic employee Jacob Coxon went viral and were echoed by others in the field who called for a slowdown in AI development.

The idea that future AI tools or agents—software bots programmed to operate with some degree of autonomy—could put people at risk has been promoted online by Anthropic employees, as well as Elon Musk, who runs OpenAI, DeepMind and the AI startup xAI.

All of the employees who spoke to the BBC did so anonymously because they were not authorized to speak to the press. The BBC knows their identities.

“Is that him?”: Former OpenAI employee reacts to viral claims

“My first thought was, ‘Is that him?’” said a former OpenAI employee who knew Coxon when he worked at the company.

The person, who now works for another AI company, said he was amused by the latest wave of AI-related existential fear. He argued that proponents of the theory had provided few details to support the idea that all human lives were at stake.

The former employee said the claims were “always vague” and that, when they appeared more concrete, they often relied on major leaps in reasoning and hypothetical scenarios.

Coxon said a group of AI agents built on a currently nonexistent AI model could decide to create a biological weapon and deploy it against a target. However, he did not explain exactly how such an attack would be carried out.