The Cold War arms race between the U.S. and Soviet Union was a dangerous game of distrust, with both sides amassing enough nuclear weapons to end civilization. The turning point came in the 1980s with the advent of seismographs, satellites, and tamper-proof cameras, which allowed for verification and ultimately led to disarmament. Today, a similar dynamic is unfolding with artificial intelligence, as the U.S. and China race to develop powerful AI systems with potentially devastating consequences.
On June 12, the U.S. government restricted access to Anthropic’s Claude Mythos and Fable 5 models, citing concerns that these AI systems could be used as cyberweapons. This move has created a strategic stalemate, with neither the U.S. nor China willing to slow down development for fear of ceding victory to the other. However, leaders of major U.S. AI labs have expressed support for a slowdown, provided it can be achieved in a coordinated manner.
The Need for AI Verification Technologies
Anthropic, in a June blog post, highlighted the dilemma: “If it were possible to effectively slow the development of this [AI] technology to give ourselves more time to deal with its immense implications, we think that would likely be a good thing. But if a slowdown simply lets the least cautious actors catch up technologically, it could leave everyone less safe.” OpenAI CEO Sam Altman echoed this sentiment, calling for a new global organization to facilitate coordinated action, including the ability to slow frontier AI development when needed.
Even Vice President JD Vance has articulated a similar logic, questioning whether China would comply with a pause in AI development. “Part of this arms race component is, if we take a pause, does the People’s Republic of China not take a pause? And then we find ourselves all enslaved to PRC-mediated AI?” he said. The challenge lies in the lack of verification technologies to ensure compliance, much like the seismographs and satellites that ended the Cold War.
The Emergence of Verification Startups
A nascent group of startups is working on developing so-called “verification” tools that could facilitate restraint in AI development. These tools aim to give all sides the confidence that their rivals are complying with any agreed-upon restrictions. According to an estimate cited by Lucid Computing, one such startup, only around 50 people worldwide are currently working on these technologies. If successful, these verification tools could open up new possibilities in AI governance.
Lucid Computing’s work focuses on building upon “trusted execution environments” on specialized AI chips, developed by companies like Intel and Nvidia. These environments allow chips to compute information confidentially, even safe from the owners of the data center where they are housed. The idea is that a special piece of software could sit inside these trusted environments, examining the AI and checking for compliance with given rules, such as confirming that a specific model is being run or determining whether chips are being used in the training of a new model.
Kristian Rönn, the CEO of Lucid Computing, emphasizes the importance of avoiding the extremes theorized by philosopher Nick Bostrom, who imagined that the risks of super-powerful AI might one day incentivize states to impose totalitarian-style surveillance. “I would hate it if we had to choose between the extremes of a global pandemic [designed by AI] killing us, or a totalitarian state monitoring our every move,” Rönn says. “Both of these extremes seem really, really bad… What we’re saying is that you can actually, through cryptography, maintain privacy and have security at the same time.”
The Challenges Ahead
Getting China to trust verification tech developed in the West might prove difficult. In 2026, Beijing reprimanded Nvidia after U.S. lawmakers called for better enforcement of U.S. export controls, proposing measures that could track the locations of AI chips and even remotely disable them. If verification technologies are only developed in the U.S., there’s a risk that China will view them as spyware, which could make them essentially useless.
Across the Atlantic, a British engineering firm called Amodo Design is taking a different approach with “recomputation,” which works by re-running portions of a company’s AI workload and examining the results. Thomas Milton, who co-founded Amodo, cautions that no single verification tool is enough. Verification, he says, is “a ladder rather than a one-shot solution:” a stack of checks that can start crude and grow more rigorous over time, just like it did during the Cold War.
Both approaches carry some limitations. For example, neither Amodo nor Lucid’s methods of examining known chips could confirm that an adversary wasn’t hiding a secret data center under a mountain somewhere. “Training runs are far easier to conceal than missile silos,” Anthropic noted in June. To address this, a paper from researchers at the think tank RAND argues that at least six different types of verification may be necessary, including software monitoring systems, built-in security features in AI hardware, and more traditional measures like whistleblower protection programs and surveillance by intelligence agencies.
The path to AI governance is fraught with challenges, not least of which is the lack of agreement on what exactly needs to be verified. Slowing down in AI sounds simple in theory, but it is very hard to specify restrictions that wouldn’t leave glaring loopholes in practice. Until companies or countries get around a negotiating table, the startups trying to build verification technologies are pointing at uncertain targets.
