Someone hacked a security scanner. Then the scanner hacked everyone who trusted it. In March this year, a group called TeamPCP got into Trivy, a widely used security scanner. They used an old automation token that had been rotated but never fully switched off. That gave them about twenty days to quietly swap out Trivy's published versions for their own. This is eyebrow's entire thesis. LiteLLM was never attacked. It just had one line in its build file telling it to install Trivy, with no version number attached, no fingerprinting. So the poisoned scanner walked straight into the build, and the build shipped two backdoored releases to PyPI. They were up for forty minutes. CloudSEK reconstructed the fallout and found more than 2,500 (!) companies and around 434,000 build pipelines potentially exposed. The code ran the moment Python started, so nobody had to install or import anything for it to fire. It took SSH keys, cloud credentials, Kubernetes tokens, .env files and LLM API keys. Even the FBI was still warning about those stolen credentials in July. Our take: almost nothing here went wrong in the way people assume. The token was rotated. The scanner was reputable. The maintainers weren't sloppy. What failed was one missing version number in a CI file, the kind sitting in nearly every repo on GitHub, because pinning versions feels like paperwork until the day it isn't. And let's be honest, AI projects have this worse than most. An agent pulls in model SDKs, gateways, MCP servers, plugins and skills. Most of that is new, a lot of it floats, and plenty of it loads while the agent is running, where nothing is watching. But eyebrow always does. Your risk stopped being your own code a long time ago. It's everything your code trusts, and everything those things trust, now more than ever if you ask us. We think a lot of teams will learn this the hard way over the next year. our eyebrow fix not hot. It is boring BUT it works: know exactly what you're loading, pin it, and get told the moment it changes. We are not here for the hype but to secure the space. It might take some time for people to understand and wake up. Meanwhile, we keep shipping and growing. Our plans will bring us deeper inside AI agent territory. Hints have been shared and will be shared. But anyone that thinks we slowed down is not having things clear. We running.