ai and ml<\/p>\n
Lock it down, warns Satya Nadella, seemingly forgetting the billions Redmond chipped in to OpenAI back in the good old days<\/p>\n
Seemingly unaware of the concept of irony, Satya Nadella is warning AI-using enterprises to take care not to give away their business secrets alongside the massive piles of cash they\u2019re forking over to frontier labs every month.\u00a0<\/p>\n
Writing in a long-form<\/a> post on X over the weekend, the Microsoft CEO and chairman warned<\/a> of what he called the \u201creverse information paradox,\u201d a situation in which purchasers of AI essentially pay for the intelligence product they\u2019re getting twice: once with cash, and again \u201cwith something even more valuable,\u201d namely the proprietary business knowledge one has to feed an AI model in order to make it worth using in the rare instance<\/a> an AI investment actually pays off.\u00a0<\/p>\n \u201cOver time, the information asymmetry becomes increasingly skewed,\u201d Nadella noted. \u201cThe seller learns more and more about you as you use what you purchased, while you learn very little about what the seller is learning in return.\u201d<\/p>\n The irony is thick, given that Microsoft itself pushes AI that slurps up<\/a> business data<\/a>, and Redmond helped get this entire messy AI ball rolling by investing billions into early generative AI leader OpenAI. Azure was the former exclusive cloud home for ChatGPT, and Microsoft leadership arguably<\/a> helped Altman get his job back when OpenAI ousted him in 2023. The pair\u2019s relationship grew\u00a0strained<\/a> in the intervening years, and they\u00a0loosened<\/a> several exclusivity provisions in early 2026.<\/p>\n It also comes after a number of large organizations paused or restricted Microsoft Copilot deployments in 2024 over a related concern: weak data governance and sprawling internal access rights.<\/p>\n Enterprise data security outfit Securiti told The Register in 2024 that about half of the more than 20 chief data officers it polled had grounded Copilot deployments, either switching the assistant off or severely restricting what it could access.\u00a0The problem was particularly acute in organizations with years of accumulated SharePoint and Microsoft 365 permissions, where overly broad access rights risked exposing sensitive information through Copilot.<\/p>\n Fast forward a couple of years, and now Nadella is warning that data protection measures aren\u2019t even enough for a business to stay safe in the AI age.\u00a0<\/p>\n \u201cModels learn from \u2018exhaust,\u2019 the prompts people write, the tools agents use, and especially the corrections people make,\u201d Nadella warned. \u201cIt’s the kind of knowledge a competitor could never buy, and the kind that leaks almost imperceptibly: trace by trace, correction by correction, eval by eval.\u201d<\/p>\n Consuming intelligence through AI, Nadella added, creates more organizational intelligence. The Microsoft chief argued that the knowledge generated through those interactions ought to belong to the companies that create it.<\/p>\n \u201cEnterprises need a real trust boundary for their human capital and token capital to compound,\u201d Nadella wrote, describing his ideal solution as having \u201ca hard boundary across which nothing crosses, not even the intelligence exhaust, without consent.\u201d<\/p>\n In other words, welcome to the post-cloud era when all your AI infrastructure will come home to roost inside your own network. If you think we\u2019re exaggerating, Nadella even mentions that one of the things enterprises need to do to solve the reverse information paradox is to build their own proprietary AI learning environments \u201cwithin the tenant boundary.\u201d\u00a0<\/p>\n We asked Nadella and Microsoft whether solving the problem goes beyond good data governance, as Nadella suggested in his article, and a spokesperson told us yes, describing the matter as a structural problem with the current generally accepted model of AI business in which companies rely on hosted services. Anyone and everyone using AI for business is at risk, they explained.\u00a0<\/p>\n In addition to isolating learning environments, Nadella\u2019s X note also suggested AI-using businesses should create their own private evaluation systems and retain ownership of organizational AI memory and decouple their orchestration layer from any particular AI model, essentially creating \u201cyour own continuous learning loop.\u201d\u00a0<\/p>\n \u201cA company should be able to use a model without giving up the knowledge that makes it unique,\u201d Nadella wrote.\u00a0<\/p>\n The Microsoft spokesperson argued that agent harnesses and memory should be independent of models, and called for enterprises to have the rights to their own usage data and model outputs, echoing Nadella\u2019s comments about the irony of leading AI firms crying foul about model distillation<\/a> while reserving \u201cthe right to learn from customer usage and interaction data.\u201d<\/p>\n