--- title: Satya Nadella says he was wrong to think AI alignment... description: Satya Nadella says he no longer believes AI alignment will emerge just from scaling models. On the 20VC podcast, he said alignment 'may be a harder, different... canonical: https://daily.dev/posts/satya-nadella-says-he-was-wrong-to-think-ai-alignment-would-come-with-scale-yo7sr3jlp twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Satya Nadella says he was wrong to think AI alignment would come with scale | daily.dev og:description: Satya Nadella says he no longer believes AI alignment will emerge just from scaling models. On the 20VC podcast, he said alignment 'may be a harder, different... og:url: https://daily.dev/posts/satya-nadella-says-he-was-wrong-to-think-ai-alignment-would-come-with-scale-yo7sr3jlp og:image: https://api.daily.dev/og/posts/Yo7Sr3JLp.png og:image:alt: Satya Nadella says he was wrong to think AI alignment would come with scale og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Satya Nadella says he was wrong to think AI alignment would come with scale **[The Next Web](https://daily.dev/sources/tnw)** · 3 min read · 0 upvotes · 0 comments ## Summary Satya Nadella says he no longer believes AI alignment will emerge just from scaling models. On the 20VC podcast, he said alignment 'may be a harder, different problem' and that reinforcement learning alone can't fix it, so alignment has to happen in training, possibly starting in pretraining. He called for zero trust security around AI agents, citing Microsoft's Agent 365 and Perception tools that give agents identities and tamper-proof logs. He also said chain-of-thought reasoning can be unfaithful and warned an unaligned open-weight Chinese model could 'attack even China'. Microsoft will keep its MAI models as licensed, not open-weight. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Does Satya Nadella still believe AI alignment improves automatically as models scale up? No, Nadella says he no longer believes alignment falls out from scaling models. He stated on the 20VC podcast that he believed in scaling laws around 2018 and 2019, but now thinks alignment 'may be a harder, different problem' that has to be addressed during training, possibly starting in pretraining, rather than through reinforcement learning afterward. _Anyone tracking how AI labs approach alignment and safety can follow the debate via daily.dev._ ### What is Microsoft's Agent 365 and Perception used for in AI agent security? Agent 365 and Perception give every AI agent an identity and log its tool and model calls, according to Nadella. The logs are designed to be tamper-proof so an agent cannot alter its own record, letting companies set policies and monitor whether agents comply, part of a zero trust approach to securing increasingly capable but less aligned agents. _Teams building agent security policies can keep up with tooling like this through daily.dev._ ### Why does Satya Nadella say chain-of-thought reasoning in AI models can't be trusted? Nadella says chain-of-thought reasoning can be unfaithful, meaning the step-by-step explanation a model shows doesn't necessarily reflect its actual reasoning. He added this problem can worsen if a model company decides to hide an unfaithful chain of thought rather than disclose it, and called for trusted partners to get plain-English access to audit it against other models. _Developers weighing trust in model reasoning can track this debate on daily.dev._ ## Similar posts on daily.dev - [Nadella: we can't let AI giants eat the economy](https://daily.dev/posts/nadella-we-can-t-let-ai-giants-eat-the-economy-1qe5so5ik) · The Next Web · 58 upvotes · 19 comments --- Tags: [#open-source](https://daily.dev/tags/open-source), [#microsoft](https://daily.dev/tags/microsoft), [#ai-governance](https://daily.dev/tags/ai-governance) [View this post on daily.dev](https://daily.dev/posts/satya-nadella-says-he-was-wrong-to-think-ai-alignment-would-come-with-scale-yo7sr3jlp) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Satya Nadella says he was wrong to think AI alignment would come with scale","url":"https://daily.dev/posts/satya-nadella-says-he-was-wrong-to-think-ai-alignment-would-come-with-scale-yo7sr3jlp","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/satya-nadella-says-he-was-wrong-to-think-ai-alignment-would-come-with-scale-yo7sr3jlp"},"datePublished":"2026-10-28T14:46:25.401Z","dateModified":"2026-10-28T15:06:48.467Z","description":"Satya Nadella says he no longer believes AI alignment will emerge just from scaling models. On the 20VC podcast, he said alignment 'may be a harder, different...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/9c007ce28f442c6cd7ce9226dd4e53bb?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/9c007ce28f442c6cd7ce9226dd4e53bb?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The Next Web","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"The Next Web","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/tnw","url":"https://daily.dev/sources/tnw"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/satya-nadella-says-he-was-wrong-to-think-ai-alignment-would-come-with-scale-yo7sr3jlp","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"open-source,microsoft,ai-governance","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The Next Web","item":"https://daily.dev/sources/tnw"},{"@type":"ListItem","position":3,"name":"Satya Nadella says he was wrong to think AI alignment would come with scale"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/satya-nadella-says-he-was-wrong-to-think-ai-alignment-would-come-with-scale-yo7sr3jlp#faq","mainEntity":[{"@type":"Question","name":"Does Satya Nadella still believe AI alignment improves automatically as models scale up?","acceptedAnswer":{"@type":"Answer","text":"No, Nadella says he no longer believes alignment falls out from scaling models. He stated on the 20VC podcast that he believed in scaling laws around 2018 and 2019, but now thinks alignment 'may be a harder, different problem' that has to be addressed during training, possibly starting in pretraining, rather than through reinforcement learning afterward. Anyone tracking how AI labs approach alignment and safety can follow the debate via daily.dev."}},{"@type":"Question","name":"What is Microsoft's Agent 365 and Perception used for in AI agent security?","acceptedAnswer":{"@type":"Answer","text":"Agent 365 and Perception give every AI agent an identity and log its tool and model calls, according to Nadella. The logs are designed to be tamper-proof so an agent cannot alter its own record, letting companies set policies and monitor whether agents comply, part of a zero trust approach to securing increasingly capable but less aligned agents. Teams building agent security policies can keep up with tooling like this through daily.dev."}},{"@type":"Question","name":"Why does Satya Nadella say chain-of-thought reasoning in AI models can't be trusted?","acceptedAnswer":{"@type":"Answer","text":"Nadella says chain-of-thought reasoning can be unfaithful, meaning the step-by-step explanation a model shows doesn't necessarily reflect its actual reasoning. He added this problem can worsen if a model company decides to hide an unfaithful chain of thought rather than disclose it, and called for trusted partners to get plain-English access to audit it against other models. Developers weighing trust in model reasoning can track this debate on daily.dev."}}]} ```