We provide research, tools, training and workshops to help people talking about AI know what actually gets through to their audiences.
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.
Read moreWe 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.
Read moreWhat do different publics actually understand and misunderstand about AI? How is that shifting?
What makes AI communication backfire or fail to land? How do communicators learn from this?
Which messages and messengers are proven to work? How do we get them into wider use?
How do we build lasting public understanding of AI?
Global AI investment reached around $581.7bn in 2025 [1]. A sliver of that goes to safety. A sliver of the sliver goes to helping the public understand any of it.
The people trying to explain AI mostly work from instinct. Early evidence, including the Seismic Foundation's 2025 study [2], suggests the field's dominant framings underperform. We test before we scale.
No policy position, no campaigns, no proprietary advice. Findings go to every side of the AI conversation, which is what lets one body credibly serve both.
Built on proven methods and applied to AI while the gap is still wide open.
A small, fast, failure‑tolerant round to prove the method.
To foundations that seed independent institutes: the pivotal step toward a neutral, permanent version of this.
Especially from people who've run something adjacent and watched it work or fail.