arXiv cs.AI · Jul 28, 2026
Falling Behind Drives Unsafe Development in an Idealised AI Race Experiment
A behavioural experiment with paired participants in an idealised AI race found that falling behind an opponent increases the likelihood of choosing Unsafe development, while being ahead reduces it. The study varied maximum risk levels (10%, 60%, 90%) but found no significant effect of risk level or risk preferences on behaviour.
What happened
In a framed behavioural experiment simulating an AI race, participants repeatedly chose between Safe and Unsafe development. Unsafe development offered faster progress and higher immediate payoffs but accumulated private risk. The competitive structure was held constant while the maximum risk varied across treatments. Pre-registered hypotheses about risk levels and risk preferences were not supported. Instead, exploratory analysis revealed that strategic state—specifically, falling behind—was the primary driver of Unsafe choices.
Technical significance
The experiment used a repeated game with an uncertain time horizon, where Unsafe development provided a speed advantage but increased cumulative risk. The lack of effect from risk preferences suggests that strategic dynamics, rather than individual risk tolerance, dominate decision-making in competitive AI development scenarios.
Industry impact
Competitive pressure in AI development may lead to corner-cutting on safety, even when individual developers are risk-averse. The finding that falling behind triggers unsafe behaviour implies that perceived market position could be a stronger predictor of safety lapses than regulatory risk thresholds.
What to watch
Future research could explore interventions that alter perceived competitive standing or introduce transparency mechanisms to mitigate unsafe racing dynamics. Observing whether similar patterns emerge in real-world AI development teams under time pressure would be a key next signal.
Decision value
For AI companies, the study highlights the risk that competitive benchmarking and market positioning may inadvertently incentivize unsafe practices. Leaders should consider internal incentives and monitoring systems that counteract the psychological pressure of falling behind.