Protecting democracy isa deep-tech problem.

AI systems are already contributing to scientific discoveries and to the development of their own successors. Superintelligence is coming.

Alignment is the effort to ensure that AI remains guided by human values. Following a distinction proposed by OpenAI Chief Scientist Jakub Pachocki, the alignment of AI systems can be divided into two problems: goal alignment and value alignment.¹

A value-aligned system should act with honesty, integrity, and regard for humanity.

But at societal scale, value alignment is not merely a property of a model. It is a property of the entire system surrounding it: the model, its owner, its operator, and the institutions governing its use. This is why states developing advanced AI are skeptical not only of one another’s capabilities, but of one another’s claims to alignment. A system may be aligned in the technical sense and still embody the values and interests of a political order its rivals fundamentally reject. The values superintelligence ultimately serves will depend not only on how it is trained, but also on who controls it, what it may be used for, and who has the authority to impose limits. Owners and operators will make many of these choices, but governments will ultimately define their boundaries. They will determine whether superintelligence remains privately accessible or becomes a strategic asset, whether it may be deployed in war, and whether its development and control will be centralized through a Manhattan Project–style national program. Governments exercise this authority through law.

The highest-stakes question is therefore not only whether superintelligence will be aligned.

It is: aligned to whom?

In a democracy, the answer should be the public. For all their imperfections, democratic institutions remain the best mechanism we have developed for expressing collective sovereignty while protecting the rights of individual citizens.

Citizens may be sovereign in principle. But sovereignty is not self-executing. The will of the public must pass through institutions and be translated into legal decisions before it becomes power. Law is therefore democracy’s transmission belt. And that transmission belt is malfunctioning.

Special interests capture law through its loopholes, ambiguities, exceptions, and the countless points at which influence can be inserted into the legislative process. Some of these weaknesses are accidental. Others are introduced deliberately. They are legal zero-days: vulnerabilities known to the interests positioned to exploit them, but invisible to the public expected to live under them.

Once these vulnerabilities become law, they do more than distribute money or influence. They determine who will control superintelligence and who will benefit from the abundance it creates. A model may follow its instructions perfectly while serving a sovereign that is no longer democratic. This is how value alignment becomes alignment with special interests.

The primary instrument of those interests is law. Yet law is produced through a process far removed from security-by-design engineering. It lacks formal specification, adversarial testing, systematic verification, and reliable ways to trace how an individual change affects the whole. Amendments can conceal consequences. Provisions can interact in ways that few people understand.

The law is a field sown with ambiguity; those who sow it know where to reap the harvest.

Law must leave room for judgment, interpretation, and changing circumstances. The objective is not to replace it with formal logic. It is to distinguish necessary judgment from engineered uncertainty.

For centuries, defending that system has primarily been the work of lawyers. They remain indispensable—but they are no longer sufficient. A legal system of immense complexity, confronted by machine-powered adversaries, cannot be protected through interpretation, precedent, and procedure alone. Democracy is expressed through politics, given force through law, and must now be secured through engineering.

Engineering cannot decide what society should value. That remains the work of democratic politics. But engineering can secure the process through which those values become law. Protecting democracy is therefore a foundational deep-tech problem. We formed Thessalonians to build infrastructure for democracy. Our aim is to bring the discipline of security engineering to law: to build tools that trace legal changes, test their consequences, expose hidden vulnerabilities, and detect attempts at capture before they harden into sovereign power.

To survive the arrival of superintelligence, democracy must become capable of defending itself.