The shutdown of Anthropic Fable and Mythos, and the unglamorous daily work that keeps your software safe

You come here for film cameras and old roads, and I’ll get back to those tomorrow. But I spent nearly four decades in the software industry, and last week’s news about the government shutting down some of the most powerful AI tools in existence touched on work I’ve done personally — unglamorous but essential work that almost nobody outside the industry ever sees. Last Friday, the government ordered Anthropic to shut down access to its two most powerful AI models over security concerns. The debate that followed focused on whether the government overreached. But that debate overlooked something important: those same models had already found and disclosed 1,596 security vulnerabilities across 281 open-source software projects. Working with Anthropic, Firefox fixed 271 bugs in a single release. The tools that can enable harm can also, reflexively, help fix it — at a scale no human team could match. I know a lot about this work. It is not glamorous. Modern software is rarely built from…