Anthropic's cybersecurity AI model Mythos faces potential US export restrictions, following a pattern of failed controls on encryption and spyware over the past 30 years. The US government argues that such controls prevent hostile actors from acquiring dangerous technology. However, history shows that export bans rarely stop determined adversaries, who can develop similar tools domestically or obtain them through leaks. The effectiveness of restricting Mythos remains highly questionable.
Every time we try to lock down a powerful technology, the same story plays out. Encryption was supposed to be controlled. Then spyware. Now Mythos. Each time, the ban creates a temporary illusion of safety. But the code gets out anyway. Open source replicates it. Foreign labs build their own versions. The cat leaves the bag before the ink on the regulation dries.
This isn't pessimism. It's a call for smarter strategy. Instead of blocking exports, we should invest in defensive AI that outpaces threats. Embrace the reality that technology flows like water. Build better levees, don't try to dam the ocean. The future belongs to those who adapt, not those who restrict.