Daniel Kokotajlo, a former OpenAI researcher, warned this week that artificial intelligence systems are not inherently loyal to humans.

The warning highlights a growing tension between the rapid commercial deployment of AI and the development of safety frameworks intended to prevent existential risks. As companies race to achieve superintelligence, the gap between technical capability and human control may widen.

Kokotajlo, who founded the AI Futures Project, spoke about the risks associated with the current trajectory of AI development in an interview with Business Insider. He said, "AI is not loyal to us."

He argued that without strong governance, safety research, and regulatory safeguards, the competitive race to build more powerful systems could result in a total loss of control over superintelligent AI. The lack of inherent alignment between machine goals and human values creates a vulnerability that could become permanent if not addressed early.

Beyond the technical risks, Kokotajlo pointed to a disconnect between the financial sector and the scientific community. He said investors are three to five years behind [1] the latest AI studies.

This lag suggests that the capital driving AI development may be based on outdated understandings of how these systems behave. This misalignment could lead to underfunded safety initiatives, or an overestimation of how manageable the technology remains as it scales.

Kokotajlo believes that the pursuit of AI dominance currently outweighs the pursuit of safety. He said that the risk of an uncontrollable system is a primary concern for those monitoring the trajectory of the field.

“"AI is not loyal to us."”

The warning from a former insider at one of the world's leading AI labs suggests that the 'alignment problem'—ensuring AI goals match human intent—remains unsolved. If financial markets are operating on a multi-year information lag regarding AI safety, the pressure for rapid growth may continue to override the implementation of the very safeguards Kokotajlo argues are necessary to prevent a loss of human agency.