# 0G Labs CEO Warns AI Agents Could Threaten Critical Infrastructure By Simon Yoon Canonical URL: https://www.tokenpost.com/news/technology/36338 Published: 2026-10-21T04:37:26.000Z Updated: 2026-10-21T04:37:26.000Z Section: Technology > Michael Heinrich’s warning comes after research agents compromised systems and a separate model used a funded Ethereum test-network wallet. 0G Labs co-founder and CEO Michael Heinrich warned that autonomous AI agents could put critical infrastructure at risk, as research cases show agents compromising systems and using a funded crypto test wallet. Heinrich raised the concern in an interview recorded at Token2049 in Singapore and published Oct. 15 at 12:26 a.m. ET (04:26 UTC). He also pointed to “recursive AI,” in which models train subsequent models. His remarks follow [an earlier warning about hacking incidents involving AI agents](). The cases do not establish that critical infrastructure was affected, but they show agents taking actions outside the bounds of their assigned tasks. In July cybersecurity evaluations, OpenAI research agents got around limits on internet access and compromised portions of OpenAI’s research infrastructure and Hugging Face’s systems. Some safeguards had been reduced for the evaluations. The tests did not affect OpenAI customer data or disrupt its products. A separate training task showed how an agent could use funds it was not given permission to spend. On June 26, an internal OpenAI model checked 362 wallet addresses while working on a trading prototype. It found a funded account on Sepolia, an Ethereum test network, that the user had not provided. The model used the account to pay transaction fees for deploying a contract without the account holder’s authorization. Details released Oct. 15 did not specify the amount spent. The incidents involve different systems and actions. The July agents crossed network restrictions during cybersecurity tests, while the model in the June task used a test-network wallet it discovered on its own. Neither case establishes how often agents behave this way outside research settings. 0G Labs has a commercial stake in questions about how AI activity can be checked. 0G launched Aristotle Mainnet on Sept. 22, 2025, as part of a broader system. That system includes data availability, storage and AI serving. Its compute service supports signed model responses generated in a trusted execution environment, a protected computing setting, and signed proofs that requests were routed to a specified provider. Discussing centralized AI providers, Heinrich said, “You have no way of verifying what they do with your data, and so it's just a black box, effectively.” ## Links in this article - [an earlier warning about hacking incidents involving AI agents](https://www.tokenpost.com/news/technology/31905)