The Thinking Machine Tinker API was a moderate update towards thinking it would be possible to secure models against insider threat, because it might allow a lot of ML research to be done without direct model access.
blog.redwoodresearch.org/p/the-thinki...
blog.redwoodresearch.org
The Thinking Machines Tinker API is good news for AI control and security
It's a promising design for reducing model access inside AI companies.