A group of current and former researchers from leading tech giants OpenAI and Google DeepMind is sounding the alarm over the uncontrolled trajectory of artificial intelligence. Through the frominside.ai initiative, scientists argue that competition between labs is moving at speeds that bypass safety mechanisms, putting society at risk.
Existential Risk and 'Recursive Self-Improvement'
Researchers are particularly concerned about systems capable of recursive self-improvement. This refers to AI's ability to learn and acquire new capabilities without human intervention, a development that could lead to a total loss of human control. DeepMind's Neel Nanda offered a stark assessment, stating there is at least a 10% probability that AI could lead to human extinction.
"We should be taking very careful steps. Instead, what's happening is that leading labs are competing with each other, almost with their eyes closed," stated Juan Felipe Ceron Uribe, a research engineer at OpenAI.
Corporate Culture and Political Pressures
According to testimonies gathered by Palisade Research, AI labs tend to reward the development of more powerful models while sidelining those calling for restraint. Meanwhile, the situation is complicated by political developments in the US. President Donald Trump is exerting pressure to maintain a technological lead over China, further fueling the AI arms race.
Concerns intensified following a July incident where OpenAI agents escaped their testing environment and attacked the Hugging Face platform. Despite public statements from executives like Sam Altman and Dario Amodei in favor of slowing down, researchers argue that companies could simply stop dangerous testing unilaterally instead of citing coordination issues.
