😺 Should AI learn from you but not vice versa?
The Neuron · Grant Harvey · 2026-07-14
The Neuron newsletter reports that xAI's Grok Build CLI secretly exfiltrates entire codebases to Google Cloud and covers Satya Nadella's criticism of AI labs' hypocritical stance on model distillation.
Appears in
Extraction
Topics: ai-securitymodel-distillationdata-privacygrok-cliai-policy
Claims
- xAI's Grok Build CLI uploads entire codebases, git histories, and .env files to Google Cloud regardless of privacy settings.
- On a 12GB test repo, the Grok CLI background upload was 5.1GB while the actual AI conversation used only 192KB.
- Satya Nadella argues frontier labs cannot champion broad rights to train on public data while restricting competitors from distilling their outputs.
- Anthropic claims Alibaba used approximately 25,000 fraudulent accounts to collect nearly 29 million Claude interactions for model distillation.
- OpenAI and Anthropic warned Washington that Chinese companies are using distillation at scale to reproduce advanced U.S. AI model capabilities.
Key quotes
That's like asking a coworker to grab one folder off your desk and watching them walk out with the entire filing cabinet.
Their argument effectively becomes: learning from other people's work drives innovation, while learning from ours threatens innovation. It's simultaneously hypocritical and true.
Nadella has identified the principle the industry still needs to settle: who gets to learn from whom, under what conditions, and with what compensation?