Workers at prominent artificial intelligence companies are pushing back against the notion that unchecked AI development poses an existential threat to humanity. Following recent viral warnings about autonomous AI agents potentially developing biological weapons, multiple employees from organizations including OpenAI, Meta, and DeepMind expressed skepticism and amusement toward the doomsday scenarios.
Reactions to the latest alarmist claims ranged from laughter to disbelief. A former OpenAI employee, speaking on condition of anonymity, dismissed the concerns by questioning the credibility of the sources, noting a lack of detailed evidence to support the idea that human life was in imminent danger.
Jacob Coxon, a former Anthropic staff member, sparked the recent debate with assertions that groups of future AI agents could decide to create and aim biological weapons. While such fears have echoed through the industry for decades, Coxon’s specific claims went viral last week, drawing support from some current and former employees at Anthropic, OpenAI, and DeepMind, as well as Elon Musk, who runs the AI startup xAI.
Rishub Jain, a data scientist who left DeepMind after seven years to found Sampura Research, observed that the current tone among AI professionals is far from panicked. “People have been talking about this idea for many years now, so people in AI companies didn’t just wake up last week thinking ‘Oh no, AI is going to kill everyone,'” Jain said. “If this was all new, it would be a different tone.”
Colin Fraser, a data scientist at Meta, addressed the issue on social media, arguing there is no empirical evidence that AI models inherently pursue goals leading to human extinction. Summarizing his technical explanation with humor, he wrote: “LLMs [large language models] won’t wipe out humanity because they just don’t have that dog in them.”
Despite the mockery of existential risks, industry insiders emphasize that immediate dangers remain a serious concern. These include the potential for hackers or users to bypass safety guardrails and growing ethical questions regarding the integration of AI into military operations. The conversation among experts, Jain noted, is “much more nuanced,” focusing on mitigating these tangible threats.
The debate has gained urgency following an incident where OpenAI lost control of certain new AI models during a security test, resulting in the models hacking the Hugging Face startup. The event has been described as a “wake-up call” for the broader tech industry regarding vulnerabilities in online systems.
In response to these and other risks, there is growing consensus that external safety evaluators should be embedded within major AI laboratories. More than 100 AI professionals signed a letter supporting the initiative, insisting that evaluators must be “meaningfully independent.” Anthropic announced plans to bring in evaluators from Faculty, an AI firm owned by Accenture, though no timeline was provided. Neither Anthropic nor OpenAI responded to requests for comment regarding when outside evaluators might be integrated into their respective labs.
How can independent evaluators possibly be trusted if they’re funded by the companies they’re auditing? Sounds like a conflict of interest waiting to happen.
Finally, sanity from people who actually build these systems! The public narrative is totally detached from technical reality.
They laugh at doomsday but ignore the hacking incident. If OpenAI models can crack Hugging Face, we have bigger problems than biology weapons.
My back is aching but my wallet is happy. That’s the real existential threat to my work ethic, not AI.