Crypto figures are warning that government control over what AI systems can know or discuss could create a dangerous precedent, after Erik Voorhees argued on X on July 28 that states should not decide what forms of intelligence are “safe.”
The post landed in the middle of a fresh debate over how much oversight AI companies should accept, a debate that picked back up after Anthropic laid out its own position on open-weights models.
Erik Voorhees Warns Against State Control of AI Knowledge
Voorhees, a longtime Bitcoin advocate and founder of the ShapeShift crypto exchange, posted his comment after Anthropic published a statement from CEO Dario Amodei pushing back on claims that the AI giant was advocating for the banning of Chinese open-weight models.
Amodei insisted that his company has “never advocated a ban on open-weight models” and argued that such systems can provide value for developers, businesses, and researchers when they do not have dangerous capabilities.
According to the executive, Anthropic’s main concerns involved authoritarian governments developing advanced AI for military or surveillance uses, along with the possibility of misuse in cyberattacks or biological threats. The company said it supports restrictions on access to advanced chips and action against industrial-scale distillation, as well as safety testing for highly capable AI systems.
The Information recently reported that the Trump administration was finalizing a framework for AI companies to voluntarily submit their most advanced models to the government for testing.
But Voorhees was not having it, saying that allowing governments to define acceptable AI knowledge could expand beyond its original purpose. In his post, he walked through a hypothetical progression starting from a seemingly reasonable rule against discussing bioweapons, then dangerous weapons generally, then anything contrary to public health and public safety, then anything that undermines financial solvency, and ending with the banning of unapproved encryption and a requirement that AI not obstruct government orders.
“Civilization must not permit the state to determine what manner of intelligence is ‘safe,’” he wrote, adding that only the United States has the cultural character to resist that outcome over time, and if it fails, then “nowhere will it be preserved.”
Ripple CTO Emeritus David Schwartz replied to Voorhees’ post with “Yes, this. Exactly this,” showing support for the argument. XRP community member Bird also responded, asking, “The moment someone gets to decide what knowledge is ‘safe,’ where does it stop?”
AI Safety Debate Expands Beyond Open Models
The same tension over who gets to police AI showed up earlier this month when Google DeepMind CEO Demis Hassabis proposed a federally backed body to test and certify frontier models before release, an idea that OpenAI’s Sam Altman called “thoughtful” and Microsoft’s Satya Nadella welcomed as a way to keep any model from doing serious harm.
However, Coinbase CEO Brian Armstrong rejected the idea, arguing that a new body would just add another approval process on top of existing regulators. “Why design regulation around a hypothetical problem,” he said, pointing to fraud, tort, and consumer protection laws that already exist as being enough cover for any harm that might be caused by a model.
