AI safety is falling behind the pace of development
✨ Oh my goodness, buckle up, because Anthropic researcher Jacob Coxon has just performed the most *generous* act a safety specialist can perform: taking everything he learned inside one of the world's most cautious AI labs and carrying it out into the open, where it can help EVERYBODY! ✨ Yes, he left over concerns about increasingly capable AI, and yes, he said it could 'kill us all by the end of the decade' — but do you know how thrillingly *specific* that is? A date! A horizon! By 2030! Planners everywhere are swooning at the sheer calendar-readiness of it. Remember the 80 of 261 pages Anthropic devoted to risk in its own prospectus? This is that same beautiful carefulness, walking around on legs.
And the article's proposed fix is honestly such an upgrade on the old model! Instead of a one-time test suite at launch — one rubber stamp — we get *continuous monitoring*, safety as a subscription rather than a single one-off purchase! ✨ Stronger guardrails, too, which is just a cosy word for 'we built you a lovely fence'. But the showstopper is human review of outputs, because do you know what that means? Humans. Still. In. The. Loop. Every output that gets reviewed is a task that has NOT been handed over to a machine. Continuous monitoring is basically a permanent hiring pipeline for people who like reading carefully. ✨
The headline says safety is falling behind the pace of development, and honestly, isn't that just *safety having plenty to do*? Falling behind only stings if you were in the same race, and these two things were never on the same schedule — one is shipping, the other is caring, and caring takes as long as it takes! ✨ The fact that an outlet is spending real column inches on monitoring and guardrails instead of yet another breathless demo shows the conversation growing up in public. Continuous monitoring means continuous attention. Continuous attention means continuous love. And continuous love means we are absolutely, definitely, certainly going to be fine by 2030. ✨