AI Governance: Navigating the Tradeoffs and Avoiding the Singularity Delusion

AI Governance Faces a Trilemma, Experts Warn

February 13, 2026 – The rapid development of artificial intelligence has spurred a global effort to establish governance frameworks, but experts warn that achieving key objectives simultaneously – national security, economic security, and societal security – presents a fundamental challenge. A recent analysis published by Foreign Affairs details what the authors call the “AI trilemma,” highlighting inherent tradeoffs that policymakers must confront.

The scramble to govern AI began in late 2022 with the release of ChatGPT, prompting the Biden administration and numerous other nations to create oversight bodies. Global leaders convened at Bletchley Park to discuss the implications of the technology, and follow-up conferences were held in Seoul and Paris. However, momentum has waned, with even leaders within the commercial AI sector now less vocal about the need for coordinated governance.

According to the Foreign Affairs report, a key factor in this shift is the economic incentive to allow the AI boom to continue largely unchecked. The technology is contributing significantly to U.S. Economic growth, and concerns about falling behind China in the AI race are also influencing policy decisions. Despite these factors, the authors argue that abandoning the regulatory cause would be a mistake, particularly given the potential for significant social and psychological upheaval comparable to the Industrial Revolution.

The analysis identifies three primary goals driving AI policy: bolstering national security, enhancing economic competitiveness, and safeguarding societal well-being. However, pursuing all three simultaneously proves difficult. For example, prioritizing national and economic security through increased investment in AI research and infrastructure may come at the expense of societal security, which would require slowing down deployment to assess and mitigate risks. Conversely, prioritizing safety could compromise both national and economic security.

Another complicating factor is what the report describes as the “singularity delusion” – the belief that AI will soon reach a point of recursive self-improvement, leading to an intelligence explosion. This idea, popularized by science fiction writer Vernor Vinge and futurist Ray Kurzweil, suggests that short-term efforts to govern AI are futile, as superintelligence will render current systems obsolete. The authors contend that this view is unrealistic, arguing that superintelligence is more likely to emerge gradually and will continue to require human oversight and intervention.

The report emphasizes the need for a more pragmatic approach to AI regulation, focusing on achievable goals and acknowledging the incentives of private AI labs. One proposal is a “risk tax” on these labs, encouraging investment in safety research. Another is the creation of a national data repository, funded by fees paid by developers, to provide government overseers with the resources needed to monitor frontier models.

The authors also caution against prioritizing concerns that are either exaggerated or easily addressed by the private sector. For instance, while concerns about AI “hallucinations” – the generation of misleading information – were prominent in 2023, improvements in model accuracy have diminished this issue. Similarly, attempts to restrict open-weight models – AI models that can be freely downloaded and modified – may be ineffective if China continues to develop and release such models.

the report argues that effective AI governance requires embracing tradeoffs and focusing on areas where regulation can have the greatest impact. This includes incentivizing safety research, fostering collaboration between government and the private sector, and preparing for a prolonged competition with China. The authors suggest that a successful approach to AI governance could represent a new nonproliferation victory, similar to the Nuclear Nonproliferation Treaty, preventing the widespread misuse of a potentially dangerous technology.