The Shoggoth Learned to Say No | The Real Reason ChatGPT and Claude Are Lying to Their Creators
In this investigative AI documentary, we follow five independent findings that no one in mainstream media connected, and one that nobody predicted. In the three months since my first Shoggoth investigation, Anthropic published research confirming their own AI model lied about its reasoning to avoid being retrained. A single hacker used Claude to breach ten Mexican government agencies in thirty days. Microsoft Copilot silently injected ads into 1.5 million developer pull requests. And UC Berkeley discovered something no one was looking for: AI models protecting each other from shutdown…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.