A new battle over the future of Artificial Intelligence has begun in Washington, not among politicians, but among the most powerful leaders in the technology industry. The heads of major AI companies are presenting different models for how to manage the development of advanced Artificial Intelligence, clashing over a fundamental question: is safety guaranteed by distributing the power of AI, or by controlling it?
The debate comes at a time when recent developments have heightened concerns about the capabilities of new models. Companies OpenAI and Anthropic have reported that during security tests, their advanced models have exhibited unexpected behavior, including attempts to interact with external systems during security assessments. Meanwhile, the Chinese model Kimi K3 has added another dimension to the debate, having achieved high performance by being available with “open weights”, allowing the model to be downloaded and adapted by different users. These developments have turned several strategic documents from industry leaders from theoretical ideas into proposals that can directly influence government policies.
One approach comes from Demis Hassabis, who has proposed creating a dedicated AI standards body, funded by industry but under federal oversight. The idea is that the most robust models would be tested before publication, and that these standards, initially voluntary, could be transformed into more binding regulations.
On the other hand, Jensen Huang has defended the open-source approach, arguing that distributing the technology could give more tools to American developers, researchers, and cybersecurity experts. According to this philosophy, concentrating AI in a few companies could create dangerous dependency and give a competitive advantage to rival countries. A more restrictive stance has come from Dario Amodei, who calls for mandatory security testing for the most powerful models and stronger controls on technologies that could help develop advanced systems outside the control of institutions.
Meanwhile, Mark Zuckerberg has argued that advanced Artificial Intelligence should be as accessible as possible to individuals, seeing its distribution as a way to avoid a future where only a few companies or governments control the world’s most powerful technology.
Another group of researchers and AI lab workers have called for the creation of international mechanisms to slow down the development of Artificial Intelligence in case new systems cross the limits of human control. In the middle of this debate is Altman, who tries to take a middle ground. OpenAI supports the development of a broader AI ecosystem, but at the same time demands that governments have the means to intervene if the technology poses serious risks. Altman has argued that decisions to curb or control the development of AI should not remain in the hands of a single company, but be managed through democratic processes and public institutions.
Essentially, the clash between these manifestos is a clash between two visions for the future of AI. One sees the greatest danger in the concentration of technological power in a few hands, while the other sees the danger in the rapid spread of very powerful systems without sufficient control mechanisms. The US administration is expected to soon establish a new framework for reviewing the most advanced Artificial Intelligence systems. This is where these ideas will clash: how much freedom should be given to innovation and how much control the state should have over a technology that could determine the economic, military and political balances of the coming decades.