PARIS—The emergence of generative AI less than four years ago has already triggered a series of “Sputnik” moments.
Just as the Soviet Union’s launch of the first artificial satellite into orbit in 1957 jolted the United States into upgrading its space program, the November 2022 release of ChatGPT, a large language model displaying an unprecedented level of complexity, triggered admiration and fear around the world.
Other tech firms raced to develop similar tools, even as experts, including AI pioneers Yoshua Bengio and Geoffrey Hinton, warned that the technology could pose a “risk of extinction.”
The next Sputnik moment came in January 2025, with the release of DeepSeek-R1, a frontier model developed by a Chinese startup at a fraction of the compute cost of the US systems it rivaled. This challenged the assumption that the US held an unassailable technological lead—and the belief that controlling access to chips would slow the competition.
But perhaps the most troubling Sputnik moment was the arrival of Anthropic’s Mythos model in April. Capable of identifying vulnerabilities in financial systems, payment networks, and other critical infrastructure, Mythos seemed to mark a qualitative shift in what AI models can do.
It brings AI closer to “superintelligence”: self-learning features appear to be within reach, and with them, the limits of human control. The fact that Anthropic, rather than a governmental body, made the decision to pursue a controlled rollout (only recently, after its public release, did the US take action) underscores the glaring absence of AI governance.
The consequences could be dire. Consider EternalBlue, a software exploit developed by the US National Security Agency. Soon after it leaked in 2017, hackers used it to power the WannaCry ransomware attack on the United Kingdom’s National Health Service, forcing the cancellation of nearly 7,000 appointments. EternalBlue was also behind the NotPetya cyberattack on the global shipping giant Maersk, which caused an estimated $10 billion in damage. If that was the fallout from one leaked vulnerability, imagine what would happen if Mythos, which has already found more than 10,000 vulnerabilities, fell into the wrong hands.
The governmental response to these developments has been woefully insufficient. To be sure, international bodies have sought to guide the responsible use of AI. There has been UNESCO’s Recommendation on the Ethics of Artificial Intelligence, the Council of Europe Framework Convention on Artificial Intelligence and Human Rights, the OECD AI Principles, and the United Nations’ Global Digital Compact. These efforts, to which we contributed, recognize that there is no value in innovation that is not aligned with human rights and dignity.
So far, though, the European Union is the only jurisdiction that has turned that lesson into law, with the 2024 AI Act. But implementation is already off track. The Code of Practice, intended to help tech firms comply with the AI Act (including specific commitments for providers of models with systemic risk), is voluntary. Moreover, the Digital Omnibus has delayed the application of binding obligations for high-risk standalone systems from August 2026 to December 2027.
The AI safety summits convened by the United Kingdom, France, and India have delivered useful outcomes, particularly the establishment of safety institutes that are mandated to work together in a global network. Moreover, President Joe Biden’s 2023 executive order introduced sensible policies to ensure safety and security within AI systems.
But these efforts suffered a serious setback with the arrival of the Trump administration, which revoked the executive order and doubled down on the very incentives—profit-seeking and the scramble for geopolitical advantage—that have fueled the AI race from the outset.
The arrival of Mythos has highlighted the administration’s incoherent and shifting approach. In June, Trump signed an executive order to expand federal oversight of frontier models—a sensible step. But compliance with the pre-release framework, which calls for collaboration between agencies and developers on cybersecurity testing, is voluntary.
A few weeks later, the Department of Commerce imposed export controls on Anthropic’s latest models without clear evidence or consultation. The move was presented as a national security measure, limiting democratic oversight, alienating partners, and boosting the appeal of open-source models—a gift to China.
Moreover, the government has failed to explain why Anthropic’s models, but not OpenAI’s equally powerful ones, were targeted, inviting scrutiny as these companies line up to launch IPOs.
In all these episodes, what is striking is the lack of forward thinking. Instead of preparing for future shocks, policymakers are scrambling to respond to the last one. But frontier models’ self-learning capabilities suggest that further breakthroughs will not be gradual; instead, they will most likely arrive suddenly and have compounding effects, leaving less time to adjust than before. The voluntary frameworks, industry self-assessments, and summits that produce declarations rather than binding commitments have not prepared us for what is coming.
The original Sputnik moment triggered a genuine reorientation of priorities, institutions, and resources. The US government created NASA, increased public investment in science and technology, and 11 years later put a man on the moon.
Of course, today’s geopolitical conditions look nothing like the Cold War. But policymakers must heed the lessons of these AI developments, each more powerful than the last. To ensure that AI models—especially those that seem poised to upend labor markets and transform health and education systems—align with societal values, governments must foster democratic debate and establish meaningful guardrails. Anything less courts disaster.
Gabriela Ramos, Co-Chair of the Task Force on Inequalities and Social-Related Financial Disclosures, is a former assistant director-general for social and human sciences at UNESCO, where she oversaw the development of the Recommendation on the Ethics of AI, and a former OECD chief of staff and sherpa to the G20, G7, and APEC. Emilija Stojmenova Duh, Associate Professor of Electrical Engineering at the University of Ljubljana, is a member of the Globethics Foundation Board, a member of the EU AI Scientific Panel, and a former minister of digital transformation of Slovenia.
Copyright: Project Syndicate, 2026.
www.project-syndicate.org
