Hi, all,

Back from my conference-with-a-baby adventure in Vaasa, here’s what I’ve got for you this week:

#1 There’s so much writing about AI that it’s hard to locate the good stuff. This, to me, fell into that category: Detecting AI Agent Failures Is Not Enough to Govern Them “Recognizing harm is a first step and establishes grounds for action. Governing requires the capacity and willingness to respond.“

#2 Be careful with what you wish for: AIs as Modern Genies

#3 I have not been doing a great amount of science communication recently but I have been thinking about it a lot while listening to the podcast Shadia ja viestintuojat. Recommended to Finnish language listeners!

#4 The registration is still open for the AI for Humanity and Society and Humanity 2026 conference, in Linköping on November 10-11. I’m co-chairing so I can vouch that we have a great line-up and this’ll be worth attending!

-A