Nuno Valério

Nuno ValérioOct 9, 2026

If an AI agent starts doing something wrong in one of your systems at three in the morning on a Sunday, who stops it? A real name. How fast? I ask colleagues this question a lot. The answers are rarely quick. On 23 September, the same question reached the UN Security Council and the US Congress, on the same day. At the UN, Bengio, Altman, Amodei and Delangue warned about AI systems slipping out of human control. No agreement came out of it. In Washington, Sanders and Casar introduced a bill to ban superintelligent AI, with up to twenty years in prison for anyone who builds it. Both explained how bad AI could get. Neither explained who would notice in time, or who would actually stop it. In pharma, every risk assessment has to answer three questions. How bad? How likely? Would we notice in time? We learned the hard way that skipping the third one doesn't work. This week's Trust Architecture is about that missing third question, and what a real stop button would need. And I'm still collecting your numbers from TA #23. Hours, days, or never?

Nuno ValérioOct 2, 2026

Some of the oldest wisdom we have was simply wrong. Beat the child and he'll grow up right. Some people are born to serve. Half of humanity is lesser. For centuries, those sat next to "feed the stranger" with exactly the same authority. While writing this week's TA edition, I noticed what the failures have in common. They were claims about how the world works, usually claims that kept someone powerful comfortable. Science took them apart, fairly. What survived is about the creature we are, and that creature hasn't changed much in three thousand years. And there's a numbers side too, right? Three thousand years is roughly 120 generations of people testing ideas about how to live, and living with the results. The modern era is about eight. In 2015, when researchers tried to repeat 100 published psychology studies, only about a third held up. We know vastly more about the world than our ancestors did... and, surprisingly, little that's new and proven about how to live in it. Which makes me look at AI differently. A model receives the filtered list, the old wisdom and the new knowledge together, all at once, without a single one of those generations. A speedrun to the ending screen, glitching through every wall the rest of us had to walk around. I don't know yet if that matters. I suspect we will all know, maybe sooner than expected. Maybe the ending screen is all anyone needs. I don't think it is, but I can't prove it. TA Edition link in the first comment. Which piece of "ancient wisdom" do you still follow, that you suspect is past its expiry date?

Nuno ValérioOct 5, 2026

Marginalia #9 is in Portuguese. You know, some things only come out right if in the language you swear in. Everyone back home has an opinion on Cristiano Ronaldo walking out of the national team camp, last week. Mine took 1,500 words, and stopped being about football somewhere around the middle... It became a piece about "irreplaceables". The more irreplaceable someone becomes, the more expensive the truth gets for whoever has to tell it... until nobody around them can afford it. English translation, right below the Portuguese version.