“OpenAI just launched something on July 17 that will change your business forever: ChatGPT Agent,” shares Iterate.ai CEO Jon Nordmark. “This isn’t just another chatbot. It’s an AI that takes real actions. It can book flights with your credit card. It can read your confidential files and make decisions without asking permission.”
Run on shared servers, these agents require unprecedented access to your most sensitive data and systems and they don’t forget. These agents create long-term memory, make inferences about you and act on your behalf.
Nordmark shared that when using ChatGPT, his agent “remembered” information from chats months ago, including the names of his dogs, his medical information, including height and weight and even heart rate.
This can cause a myriad of questions about ethics and security.
“This memory isn’t neutral. It’s predictive, pattern-based, and often invisible,” Nordmark explains. “You can’t see what it remembers. Or how it’s using that memory to act on your behalf.”
Essentially, this process creates and curates a “Digital You.”
Eventually, the technology will be able to map out your day, from your coffee run to your workout schedule, to your stress levels and your sleep patterns.
Nordmark shares, “this may sound extreme, but if your calendar shows a 3PM break, your location pings near a Starbucks, and your expense log shows a $4.25 charge, the agent puts it together. When agents get access to calendars, messages, receipts, apps, and location history, they don’t see you sipping—they infer it with startling precision.”
Jon Nordmark explains the security concerns with this technology.
“In fact, agents built by all big public LLM builders—OpenAI (ChatGPT), Google (Gemini), Anthropic (Claude), DeepSeek (China), Manus (China)—all run on a massive cloud platform. Each request (i.e. prompt) is processed across hundreds (or thousands) of GPUs or TPUs to support tools, memory, and real-time actions.”
Your data is processed on the same servers as thousands of other companies. Individual users, strangers, share that processing power, too.
AI Agents don’t exist in a vacuum. They store data on these servers indefinitely and can maintain live access to your systems like email, Slack, bank accounts, CRMs, and drive storage, and even payment apps.
This poses a huge amount of risk, including hackers whose focus is solely to hijack the agent.
“When agents live on shared infrastructure, one weak link becomes everyone’s risk,” states Nordmark.
And with so many companies rushing to implement AI, this can cause sloppy and risky behaviors to become the foundation of a company’s AI infrastructure, and can put data at risk.
“As leaders, we need to think twice before encouraging employees to use memory-hoarding agents—ones that act without human judgment—inside shared-cloud environments,” concludes Nordmark.
Companies and individuals who rush to use these features that are rolling out are often doing so without understanding how the technology works, where their data is going, and if their data can even actually be deleted. Not to mention the risks that open up if and when someone hacks into a server and compromises, well, everything.
About Jon Nordmark
Before co-founding Iterate, Jon was the founder and 10-year CEO of eBags.com, which grew profitably to over $100 million in revenue with a 34% CAGR from 2000 to 2008. A pioneer in e-commerce, eBags was among the first to dropship from 500+ third-party warehouses, collect customer reviews, run A/B tests, and operate on a negative cash conversion cycle. Its focus on experimentation helped make it one of the highest-converting retail sites online, recognized by Nielsen as a Top 10 U.S. retailer for conversion rates. eBags also launched and sold 6pm.com to Zappos and managed CaseLogic.com and TUMI.com globally. Before its acquisition by Samsonite, eBags served 40 million+ annual shoppers and sold over $1.65 billion in travel products. Between eBags and Iterate, Jon invested in and advised startups.
