Bill Gates has called for stronger external oversight of artificial intelligence, arguing that reliance on the goodwill of AI creators is no longer viable.
In an interview with NBC News, Gates stated that self-regulation by the industry is insufficient. He urged Congress to enact specific laws governing AI and recommended that law enforcement and policymakers become actively involved in defining guarantees and controls. While acknowledging that this approach would introduce additional bureaucracy, Gates believes it would not significantly slow down development.
Why stronger controls are needed now
Gates noted that AI has become powerful enough to cause large-scale harm if misused. He warned that AI could trigger events resulting in up to a billion deaths and described the combination of malicious actors with advanced AI tools as more dangerous than any previous weapon.
He did not advocate halting progress. Instead, he called for mandatory rules, independent checks, and ongoing monitoring, especially in high-risk areas such as medicine, finance, government systems, and other critical infrastructure.
Reactions from industry and government
Executives from Anthropic and OpenAI have also suggested slowing the pace of AI development. Former Anthropic employee Jacob Coxon publicly accused companies of acting irresponsibly and gambling with lives by racing toward self-improving superintelligence.
House Speaker Mike Johnson has indicated he prefers to wait for concrete proposals from the companies themselves. In contrast, Mark Zuckerberg opposes industry-wide coordination and maintains that individual laboratories should decide when to slow down.
Meanwhile, states including California, Maryland, and New York have started their own regulatory initiatives and are forming expert councils to address AI governance.
Practical recommendations for organizations
Organizations using AI should maintain a registry of AI tools and connected services, review prompts and data-handling scenarios, restrict access to models and internal databases, test outputs for errors and bias, log user and system actions, and prepare incident-response plans for erroneous recommendations.
