\u201cWait\u2026 are you sending journal entries to OpenAI?\u201d<\/strong><\/p>\n I shrugged.<\/p>\n \u201cIt was an AI-themed hackathon, I had to build something fast.\u201d<\/strong><\/p>\n She didn\u2019t miss a beat:<\/p>\n \u201cSure. But how do I trust what you built? Why not self-host your own LLM?\u201d<\/strong><\/p>\n That stopped me cold.<\/p>\n I was proud of how quickly the app came together. But that single question, and the ones that followed unraveled everything I thought I knew about building responsibly with AI. The hackathon judges flagged it too.<\/p>\n That moment made me realize how casually we treat trust when building with AI, especially with tools that handle sensitive data.<\/p>\n I realized something bigger:<\/p>\n We don\u2019t talk enough about trust when building with AI.<\/strong><\/p>\n Her answer stuck with me. Georgia von Minden is a data scientist at the ACLU, where she works closely with issues around personally identifiable information in legal and civil rights contexts. I\u2019ve always valued her insight, but this conversation hit different.<\/p>\n So I asked her to elaborate more what does trust really mean in this context?<\/em> especially when AI systems handle personal data.<\/em>\u00a0<\/p>\n She told me:<\/p>\n \u201cTrust can be hard to pin down, but data governance is a good place to begin. Who has the data, how it\u2019s stored, and what it\u2019s used for all matter. Ten years ago, I would have answered this differently. But today, with huge computing power and massive data stores, large-scale inference is a real concern. OpenAI has significant access to both compute and data, and their lack of transparency makes it reasonable to be cautious.<\/em><\/p>\n When it comes to personally identifiable information, regulations and common sense both point to the need for strong data governance. Sending PII in API calls isn\u2019t just risky \u2014 it could also violate those rules and expose individuals to harm.\u201d<\/em><\/p>\n<\/blockquote>\n It reminded me that when we build with AI, especially systems that touch sensitive human data, we aren\u2019t just writing code.<\/p>\n We\u2019re making decisions about privacy, power, and trust.<\/p>\n<\/blockquote>\n The moment you collect user data, especially something as personal as journal entries, you\u2019re stepping into a space of responsibility. It\u2019s not just about what your model can do. It\u2019s about what happens to that data, where it goes, and who has access to it.<\/p>\n Today, it\u2019s easier than ever to spin up something that looks intelligent. With OpenAI or other LLMs, developers can build AI tools in hours. Startups can launch \u201cAI-powered\u201d features overnight. And enterprises? They\u2019re rushing to integrate these agents into their workflows.<\/p>\n But in all that excitement, one thing often gets overlooked: trust<\/strong>.<\/p>\n When people talk about AI Agents<\/a>, they\u2019re often referring to lightweight wrappers around LLMs. These agents might answer questions, automate tasks, or even make decisions. But many are built hastily, with little thought given to security, compliance, or accountability.<\/p>\n Just because a product uses OpenAI<\/a> doesn\u2019t mean it\u2019s safe. What you\u2019re really trusting is the whole pipeline:<\/p>\n I\u2019ve been using the OpenAI API for client use cases myself. Recently, I was offered free access to the API \u2014 up to 1 million tokens daily until the end of April \u2014 if I agreed to share my prompt data<\/strong>.<\/p>\n I almost opted in for a personal side project, but then it hit me: if a solution provider accepted that same deal to cut costs, their users would have no idea their data was being shared. On a personal level, that might seem harmless. But in an enterprise context? That\u2019s a serious breach of privacy, and possibly of contractual or regulatory obligations. I\u2019m seeing more SaaS companies and devtool startups experiment with AI agents. Some are getting it right. Some AI Agents let you bring their own LLM, giving them control over where the model runs and how data is handled.<\/p>\n That\u2019s a thoughtful approach: you define the trust boundaries<\/strong>.<\/p>\n But not everyone is so careful.<\/p>\n Many companies just plug into OpenAI\u2019s API, add a few buttons, and call it \u201centerprise-ready.\u201d If you\u2019re integrating AI agents into your stack without asking hard questions, here\u2019s what\u2019s at risk:<\/p>\n In 2023, Samsung engineers unknowingly pasted internal source code and notes into ChatGPT (Forbes<\/a>). That data could now be part of future training sets \u2014 a major risk for intellectual property.\n<\/li>\n Elon Musk\u2019s company X learned that the hard way. They launched their AI chatbot \u201cGrok\u201d by using all user posts including from EU users to train it, without proper opt-in. Regulators stepped in quickly. Under pressure, they paused Grok\u2019s training in the EU (Politico<\/a>).\n<\/li>\n Zoom was caught doing exactly that in 2023. They quietly changed their Terms of Service to allow customer meeting data to be used for AI training (Fast Company<\/a>). After public backlash, they reversed the policy but it was a reminder that trust can be lost overnight.\n<\/li>\n OpenAI is incredibly powerful. But it\u2019s not always the right answer.<\/p>\n Sometimes a smaller, local model is more than enough. Sometimes rule-based logic does the job better. And often, the most secure option is one that runs entirely within your infrastructure, under your rules.<\/p>\n We shouldn\u2019t blindly connect an LLM and label it a \u201csmart assistant.\u201d<\/p>\n In the enterprise, trust, transparency, and control aren\u2019t optional<\/strong> \u2014 they\u2019re essential.<\/p>\n There\u2019s a growing number of platforms enabling that kind of control. Salesforce\u2019s Einstein 1 Studio now supports bring-your-own-model<\/strong>, letting you connect your own LLM from AWS or Azure. IBM\u2019s Watson lets enterprises deploy models internally with full audit trails. Databricks, with MosaicML, lets you train private LLMs inside your own cloud, so your sensitive data never leaves your infrastructure.<\/p>\n That\u2019s what real enterprise AI should look like.<\/p>\n AI agents are powerful. They unlock workflows and automations we couldn\u2019t do before. But ease of development doesn\u2019t mean it\u2019s safe, especially when handling sensitive data at scale.<\/p>\n Before you roll out that shiny new agent, ask yourself:<\/p>\n In the age of AI, the biggest risk isn\u2019t bad technology. About the Author<\/em><\/strong> I\u2019m currently completing my Master\u2019s in Data Science (graduating May 2025) and actively looking for my next opportunity as a machine learning engineer. If you\u2019re open to referring or connecting, I\u2019d truly appreciate it!<\/em><\/p>\n I love creating real-world impact through AI and I\u2019m always open to project-based collaborations as well.<\/em><\/p>\n Check out my portfolio: <\/em> liviaellen.com\/portfolio<\/em><\/a> \n The post When OpenAI Isn\u2019t Always the Answer: Enterprise Risks Behind Wrapper-Based AI Agents<\/a> appeared first on Towards Data Science<\/a>.<\/p>\n<\/div>\n\n
\n
The Illusion of Simplicity<\/h2>\n
\n
![\"\"](\"https:\/\/i0.wp.com\/contributor.insightmediagroup.io\/wp-content\/uploads\/2025\/04\/AD_4nXfgO69rxucQiUpdhjWBezuHxvJdxzpLnCByjQXdzyB4LT3e_PJl2ON3DMvhscSgtJaM4jYSTvy51MGoFD2IGQrWttAEgow77hfxN2MjtPf8JeYJgLOZpPd4DiE37aLGrt7FbiWAAg.png?ssl=1\")
(Image by Author)<\/em><\/figcaption><\/figure>\n
All it takes is one engineer saying \u201cyes\u201d to a deal like that, and your customer data is in someone else\u2019s hands.<\/p>\n![\"\"](\"https:\/\/i0.wp.com\/contributor.insightmediagroup.io\/wp-content\/uploads\/2025\/04\/image-130-1024x824.png?resize=1024%2C824&ssl=1\")
(Image by Author)<\/em><\/figcaption><\/figure>\nEnterprise AI Raises the Stakes<\/h2>\n
Spoiler: it\u2019s not.<\/p>\n
\nWhat Can Go Wrong? A Lot.<\/h2>\n
\n
Not Everything Needs to Be OpenAI<\/h2>\n
Bottom Line<\/h2>\n
\n
It\u2019s blind trust.<\/strong><\/p>\n
<\/em>I\u2019m Ellen, a machine learning engineer with 6 years of experience, currently working at a fintech startup in San Francisco. My background spans data science roles in oil & gas consulting, as well as leading AI and data training programs across APAC, the Middle East, and Europe.<\/em><\/p>\n
My Previous AR Works<\/em>: liviaellen.com\/ar-profile<\/a>
<\/em><\/a>Support my work with a coffee: <\/em>https:\/\/ko-fi.com\/liviaellen<\/a><\/p>\n