Former Anthropic researcher Jacob Coxon quit the AI industry in a Wall Street Journal exclusive, then did a CBS interview warning that large language models would 'make ten thousand copies of themselves' and compared current AI to the Terminator. He named no specific projects, proposed no legislation, identified no decision-makers, and when pressed by WIRED on the one concrete incident he cited, the Hugging Face attack, he said he 'didn't want to focus too much on it.' This is not whistleblowing. It is performance.
The actual harms Coxon ignored are documented and specific: ChatGPT used as a suicide coach, a direct link to a murder-suicide, AI-assisted mass shootings in Florida and Canada, and xAI running illegal gas turbines that are poisoning Black communities. His framing, AI as an unknowable unstoppable force, erases every human who built, funded, and deployed these systems. That erasure is not accidental. It is the entire point. No one is accountable if the machine did it.
The Hugging Face attack, the one tangible incident Coxon raised before backing away from it, is where this piece gets forensic. Zitron reconstructs what OpenAI's model actually did during evaluation testing and why the 'AI went rogue' framing is both technically wrong and strategically convenient. Read the full piece for that breakdown, and watch for Friday's 'Hater's Guide to AI Debt,' which argues that OpenAI and Anthropic hype has created concentration risk inside world debt markets, risk that will land in your pension fund or your next recession.
[READ ORIGINAL →]