Fixing the Game Podcast by Luke
#48 Decentralising AI: Privacy, Autonomy & the Future of Human-Centered Tech with Andrew Sispoidis
Every time you use ChatGPT, that query gets tied back to you, even if you're running it "locally," even on an enterprise account. In this episode, Luke talks with Andrew Sispoidis, founder and CEO of ESMOS AI, about why today's AI companies can't avoid retaining your data, and what it would actually take to build an AI gateway that's private by design, not by policy. Andrew explains why centralization is the root problem, large language models are built to take, analyze, and retain data and how ESMOS strips the "who" from the "what," so a query can be answered without ever being tied back to the person who asked it. He gets specific about the risks most people never think about: how anonymized data gets re-correlated back to individuals, how insurers can use query patterns to adjust rates, and why running a model "locally" still leaves an audit trail on someone else's servers. We also get into the product itself — how ESMOS routes queries through multiple encrypted gateways, strips metadata down to a single timestamp, and lets users chain a question across multiple AI models without ever surfacing their identity — plus why professionals in regulated industries (legal, healthcare, finance) are becoming the platform's earliest and most natural users. Key topics: Why centralized AI companies structurally can't avoid retaining your data How "anonymized" data gets re-identified, and real examples of the fallout (including insurance rates) Why running an LLM locally doesn't actually solve the privacy problem How ESMOS separates identity from content across every query Why regulated professionals are the first real market for private-by-design AI Learn more about Andrew's work and ESMOS AI, and how they're building toward a fully decentralized, private gateway to large language models.
