Mallory Stewart: Treaties and agreements are still something we should work toward -- risk reduction -- but we should look to new methods to compel responsible behaviors. #CACNP2025

@joecirin.bsky.social: 7 of the 9 nuclear states are under the control of one authoritarian leader. That's the biggest challenge we face. #CACNP2025

@laicie.bsky.social @joecirin.bsky.social: This new reality requires new thinking: old theories don't really fit what we're in right now. We're in a new nuclear arms race, the first since the 1980s... our goal has to be to minimize those losses and prevent the worst from happening.

Dr. Cindy Vestergaard on AI creating disinfo: Mis and disinfo is not a new thing in the nuclear sector (Fukushima)... but what does it mean if you are trying to do threat reduction, when the physical and digital lines are getting more merged and reality is more blurred? #CACNP2025

Dr. Cindy Vestergaard: Humans have emotions. Humans have reasoning... When we're looking at potential for IAEA to use AI, there is potential but we have a long way to go for testing, demonstration, testing again, certification & some of those certification instruments take 7-10 years. #CACNP2025

@jackieschneider.bsky.social: The problem as we move toward more AI-enabled early warning is that the user doesn't nec have the ability to interrogate truth behind AI. Too many censors, the model is too complicated & that leads to a few incentives for both accidental & inadvertent escalation.

@jackieschneider.bsky.social: Keeping a human in the loop in nuclear decisions is such a resoundingly logical decision...but the devil is in the details... AI can mimic human behavior but it has significant bias, and that bias tends toward escalation. #CACNP2025

Dr. Jesse Kirkpatrick: If I have to have two-step authentication on my phone, we should have it in a system such as this (nuclear command, control and communications) #CACNP2025

Dr. Jesse Kirkpatrick: We have an AI system that can result in confident but catastrophically wrong consequences... But AI doesn't reason. It can be precise, persuasive and completely wrong in a way that looks like it's completely right. #CACNP2025

@mjgault.bsky.social: We have avoided nuclear war because we've been lucky, but also because we've been human... The real danger of nukes at the dawn of AI is that justified by efficiency in the name of modernization, AI will replace the next Stanislov Petrov