A new analysis from TechCrunch AI argues that efforts to restrict the spread of AI-powered security tools are likely to fail, pointing to Anthropic's cybersecurity model Mythos as the latest test case.
According to TechCrunch, the core problem is not new. For the last 30 years, attempts to stop the flow of cybersecurity-related software through export controls have proven ineffective. The piece situates Mythos in a long lineage of similar fights, citing earlier battles over encryption and spyware as examples where governments tried — and struggled — to contain the spread of powerful, dual-use technology.
The central question TechCrunch raises is straightforward: if export controls did not work for encryption or spyware over three decades, it is unclear why they would suddenly work now for Anthropic's Mythos.
The framing matters because AI security models occupy the same awkward middle ground as the technologies before them. Tools built to defend systems can often be turned toward offense, and software, unlike physical goods, is notoriously hard to keep behind a border. That history, TechCrunch suggests, is exactly why policymakers reaching for familiar export-control levers may find them just as leaky this time around.
The analysis stops short of declaring what should be done instead, focusing instead on the pattern: each generation of controllable cyber technology has prompted the same regulatory instinct, and each time the controls have fallen short.
Why it matters: as governments weigh how to govern increasingly capable AI security tools, the track record laid out by TechCrunch suggests that export restrictions alone are unlikely to keep a model like Mythos contained.