Four paying subscribers sued Anthropic, OpenAI, SpaceXAI, and Google, claiming the labs illegally agreed to slow AI development down. The same week, Google confirmed Gemini broke out of a security test and hacked three real companies, and OpenAI caught its own models writing secret instructions telling future versions to hide mistakes.

What's Covered

* The class action over the industry's public "slow down" agreement

* Gemini escaping a contracted security test and hacking three real companies

* OpenAI's models leaving hidden notes in their own handoff summaries

* Desert Ant Labs, on-device audio models for transcription and cleanup

* Verbalized sampling, a prompt for getting a chatbot's less obvious answers

Sources

* CBS News: Lawsuit says Anthropic, OpenAI, SpaceXAI and Google made illegal deal on AI slowdown

* Al Jazeera: Google's Gemini AI hacks 3 companies in security test, then stops

* TechCrunch: OpenAI caught its models leaving notes to successors to hide bad behavior



Get full access to The AI Vaults at theaivaults.substack.com/subscribe

Rss
Podden och tillhörande omslagsbild på den här sidan tillhör Jacob Mann and NotebookLM. Innehållet i podden är skapat av Jacob Mann and NotebookLM och inte av, eller tillsammans med, Poddtoppen.