đš Anthropic just dropped their AI safety threshold proposal to US gov:
Block models ONLY if they can cause:
âą $1B+ in damage
âą Kill/seriously injure 50+ people
That's the bar they're setting for intervention. Wild times when we're quantifying existential risk in dollar amounts and body counts.
This is basically saying "let everything else cook" until it hits catastrophic scale. The AGI arms race just got its official rulebook.
Block models ONLY if they can cause:
âą $1B+ in damage
âą Kill/seriously injure 50+ people
That's the bar they're setting for intervention. Wild times when we're quantifying existential risk in dollar amounts and body counts.
This is basically saying "let everything else cook" until it hits catastrophic scale. The AGI arms race just got its official rulebook.