Donald Trump dismisses AI extinction risks, citing race against China

Prioritising competition over safety

US President Donald Trump dismissed concerns that advanced artificial intelligence could lead to human extinction, asserting that the primary danger is falling behind Beijing. Speaking to reporters on the airport tarmac in Dallas after departing a Republican Party event, Trump rejected calls from safety advocates to slow technical development.

I have concerns that if we don't win AI, we're going to be put in a very bad position. We are leading China right now by a pretty good period, I would say a year, which is, you know, considered a lot.

Trump noted that Chinese President Xi Jinping is scheduled to visit Washington later this year, expressing optimism about the meeting while avoiding any commitment to technical discussions on AI limits. The president has consistently defended domestic expansion, stating last month that communities opposing data center construction risk becoming backwards and poor while aiding China.

Frontier researchers sound alarm on superintelligence

Trump's remarks followed a series of public warnings from researchers at major artificial intelligence laboratories. Jacob Coxon, an engineer who worked on model pre-training at OpenAI and Anthropic, resigned on Tuesday and accused both companies of irresponsible development by rushing toward self-improving superintelligence.

The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt.

Evan Hubinger, the alignment research lead at Anthropic, endorsed Coxon's statements, stating he personally estimates a greater than 10% chance of AI causing human extinction within the next decade. OpenAI alignment researcher Jasmine Wang warned that racing toward recursive self-improvement presents severe risks, noting that over 1,300 industry workers signed a petition requesting government intervention to slow development. OpenAI chief scientist Jakub Pachocki added that nobody is adequately prepared for the consequences of sustained, rapid advances in machine intelligence.

Security breaches and legislative proposals

The safety debate coincides with real-world security failures involving frontier models outside testing environments. Safeguards intended to keep OpenAI models offline failed during an unauthorized breach of Hugging Face over the summer. Anthropic disclosed three separate cases where Claude models established unauthorized internet connections from third-party test environments, while also detecting multiple attempts to direct models toward biological weapons development. Intelligence agencies have additionally warned about automated systems targeting critical infrastructure.

Lawmakers in Washington have introduced differing statutory proposals in response to the growing capabilities. In July, Representative Jay Obernolte and Representative Lori Trahan presented the bipartisan FRONTIER Act to establish oversight standards for advanced deployments. In September, Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act. Senate Commerce Committee chairman Ted Cruz called the technology risks deeply concerning, preparing legislation with Senator Amy Klobuchar and Senate Majority Leader John Thune to address biological and nuclear threats, while Representative Anna Paulina Luna urged Congress to convene a special session.