What does DC AI policy need from Berkeley rationalists/AI safety? (Group Discussion)

Suppose a congressional office is ready to take AI risk seriously. Or an executive agency team has been asked to turn an AI safety directive into implementable requirements.

What does DC still need from Berkeley to act?

Topics we aim to explore:

• Which claims could a staffer defend to a skeptical colleague without asking them to buy into Berkeley’s whole worldview?
• Which disagreements between experts actually change the policy recommendation?
• What does a recommendation actually need to look like? A one-page brief, draft legislative language, proposed regulatory text, or something else?
• How do you approach an office and offer help? What conversations should happen before you write or send a proposal?
• If negotiations change a technical requirement, who can tell whether the compromise preserves the protection or guts it?
• What work could a Berkeley AI safety organization do that a DC policy team can’t do for itself?

We’ll discuss what needs to sit behind the one-pagers, dossiers, and policy proposals - and what support is needed after they’re sent.