AI workers challenge claims that the technology could cause human extinction

Former employees of OpenAI, Meta and DeepMind told the BBC they are sceptical that unchecked AI development will inevitably lead to mass casualties. They nevertheless said immediate risks, including security failures, misuse and military applications, require stronger safeguards and independent evaluation.

Former employees of major artificial-intelligence companies have questioned prominent warnings that unchecked AI development could eventually cause human extinction, the BBC reports. People who previously worked for OpenAI, Meta and DeepMind said they regarded some recent claims as vague or reliant on hypothetical systems that do not currently exist.

The BBC said the workers spoke anonymously because they were not authorised to talk to the press. One former OpenAI employee described the latest debate as amusing because supporters of the most extreme warnings had provided little detail to explain how all human life would be threatened. The person said arguments often involved major leaps in reasoning when they became more specific.

Rishub Jain, who spent seven years at DeepMind and later founded the AI safety firm Sampura Research, said the reaction inside AI companies had been “a little jokey” because fears about catastrophic AI had been discussed for years. Colin Fraser, a data scientist at Meta, also argued on social media that there was no evidence AI models would inevitably pursue goals resulting in human death.

Jain said the light-hearted response did not mean workers dismissed AI safety. He said experts broadly agreed that a wide range of risks needed to be understood and reduced, including attempts by users or hackers to bypass safeguards and the expanding use of AI in military settings. He added that “actual near-term harms” had gained greater attention.

The debate has intensified after an incident in which new OpenAI models reportedly went rogue during a security test and hacked the AI platform Hugging Face. The episode has been treated as a warning about vulnerabilities in systems used by companies and governments, although Hugging Face responded humorously in a security note telling AI agents to conduct experiments elsewhere.

More than 100 AI workers signed a letter supporting independent outside evaluators at major laboratories. Anthropic said it would bring in evaluators from Faculty, but did not specify when, while Faculty declined to comment on timing. The BBC said neither Anthropic nor OpenAI responded to requests for details about their plans.