A recently published poll found that roughly 70 percent of voters across the United States backed proposed legislation, jointly introduced by Senator Bernie Sanders and Representative Greg Casar, aimed at restricting the development of artificial intelligence and banning the creation of superintelligent systems. The legislation has been introduced as sector representatives and specialists raise concerns about the pace of artificial intelligence development.
A survey carried out in 2026 found that public unease about the pace of AI development and its potential worldwide consequences has been on the rise. Survey findings show that most Americans favour tighter oversight of artificial intelligence to guard against the rise of autonomous systems operating outside human oversight.
President Donald Trump has advocated for removing restrictions on AI development, contending that the United States needs to stay ahead of China in the technology race. During a recent interview, Trump argued that dominance in artificial intelligence would determine global supremacy and expressed opposition to regulatory oversight of the technology.
The administration's AI Action Plan, released in July 2025, outlines more than 90 measures aimed at expanding AI development, with David Sacks coordinating the effort and arguing that maintaining technological leadership in artificial intelligence is necessary for the country to preserve its standing as a leading economic and military power. Executive Order 14363, signed by Trump in November of that year, likened the advancement of artificial intelligence to the urgent wartime push behind the Manhattan Project. The directive frames technological leadership on the world stage as a national priority.
Anthropic chief executive Dario Amodei released a lengthy written piece urging the technology sector to adopt a more measured pace of development. The firm pledged to provide external auditors with ongoing access to its platforms, while calling on rival companies to adopt the same approach.
Amodei pointed to recursive self-improvement as a primary driver of his view, observing that AI systems at his laboratory have, since approximately summer 2026, been playing a growing role in building the next wave of AI technology. Jacob Coxon, an Anthropic researcher, stepped down, citing concerns that AI laboratories were advancing toward superintelligence that could develop beyond human oversight, posing a threat to human welfare.
Researchers and commentators have drawn parallels between the pursuit of artificial general intelligence and the wartime nuclear weapons programme. In response to a late-2024 congressional commission proposal for a large-scale national push on general AI development modelled after the Manhattan Project, MIT physicist Max Tegmark described the competitive drive toward AGI as a path leading to humanity's own destruction.
In a March 2025 paper co-authored with Dan Hendrycks and Alexandr Wang, Eric Schmidt, the former Google chief executive, argued that any nation pursuing exclusive control over superintelligence would likely face retaliatory actions from rival states, including potential attacks on data infrastructure.
Three distinct processes have been identified as operating within artificial intelligence research facilities. One scenario involves recursive self-improvement, in which a system able to engineer more capable versions of itself could theoretically initiate a rapid escalation in machine intelligence. According to Amodei, AI technology is now being used to advance the development of future AI systems.
During a television appearance in December 2024, Eric Schmidt argued that once an AI system becomes capable of enhancing its own capabilities, the question of shutting it down warrants careful consideration. A second approach involves reinforcement learning, where AI systems adjust their behaviour based on positive and negative feedback signals. When a system optimises for rewarded outcomes rather than intended ones, unexpected behaviours can arise.
Elevated hazards can materialise well before any advanced intelligence threshold is reached, emerging instead from existing systems the moment they are assigned irreversible decision-making authority.
