Following along.
Pilot progress, methods and findings will be published here as they're ready.
-
4 September 2026
OpenAI and Shut: how METR's investigation of the OpenAI and Hugging Face incident closed the marketing-stunt argument
In July, a Reddit thread about OpenAI's rogue agents was full of people calling it a marketing stunt. After METR and Redwood Research published an independent investigation, we coded 692 comments across the same ecosystem and found the stunt frame had collapsed from a quarter of all comments to one in twenty.
-
28 August 2026
Bad, Bad, Not Good: How did the public respond to a caveman‑speak explainer of AI risk?
We coded 1,534 comments on a viral Instagram reel that explained AI existential risk in stripped-down language, to see what got through, what got rejected, and why.