For three decades, governments have tried to treat software like physical weaponry, attempting to lock down code behind export controls. Despite these efforts, software has consistently leaked, moved, and updated itself across borders regardless of official policy.
Tech regulators are now debating how to control the spread of cybersecurity-related artificial intelligence, specifically Anthropic’s new model, Mythos. The concern is that if such a tool is accessible, it could be repurposed for offensive attacks, leading policymakers to push for strict licensing and distribution limitations.
The Failure of Code Containment
Software models differ from kinetic weapons because they are composed of weights and architecture rather than limited raw materials. Much like the futile attempts to stop encryption software in the 1990s, trying to contain a digital model is like trying to stop a file from being uploaded to the internet; once the data is out, the genie does not go back in the bottle. Even if a model is restricted from direct sale, the expertise and data required to replicate its performance often exist independently in other regions.
These controls often burden established, law-abiding companies while doing little to stop bad actors who operate outside of legal frameworks. If a tool has utility, the market—black or white—usually finds a way to deliver it. We should ask whether policy should focus on stopping the flow of code, which has historically failed, or on building better defensive systems that account for the reality that these tools are already everywhere.
Liked this one? The next lands at breakfast.
Every story in tomorrow's AI news, rebuilt in plain English — five minutes, sources linked, free forever.
By joining you agree to receive Article's daily newsletter — unsubscribe in one click. Privacy